Observe the following precaution when handling the battery charger If an MH-71 battery charger is provided with the camera, observe the following precautions: Keep dry. Failure to observe this precaution could result in fire or electric shock. Dust on or near the metal parts of the plug should be removed with a dry cloth. Continued use could result in fire. Do not handle the power cable or go near the battery charger during thunderstorms. Failure to observe this precaution could result in electric shock. Do not damage, modify, forcibly tug or bend the power cable, place it under heavy objects, or expose it to heat or flame. Should the insulation be damaged and the wires become exposed, take it to a Nikon-authorized service representative for inspection. Failure to observe these precautions could result in fire or electric shock. Do not handle the plug or battery charger with wet hands. Failure to observe this precaution could result in electric shock. Use appropriate cables When connecting cables to the input and output jacks, use only the cables provided or sold by Nikon for the purpose, to maintain compliance with product regulations.
Handle moving parts with care Be careful that your fingers or other objects are not pinched by the lens cover or other moving parts. CD-ROMs The CD-ROMs supplied with this device should not be played back on audio CD equipment. Playing CD-ROMs on an audio CD player could cause hearing loss or damage the equipment. Observe caution when operating the flash Using the flash close to your subjects eyes could cause temporary visual impairment. Particular care should be taken if photographing infants, when the flash should be no less than one meter (39in.) from the subject. Do not fire the flash while it is touching a person or object Such use can result in the person being burned, and/or their clothes igniting from the heat of the flash. Avoid contact with liquid crystal Should the monitor break, care should be taken to avoid injury due to broken glass and to prevent liquid crystal from the monitor touching the skin or entering the eyes or mouth.

Notices

No part of the manuals included with this product may be reproduced, transmitted, transcribed, stored in a retrieval system, or translated into any language in any form, by any means, without Nikons prior written permission. Nikon reserves the right to change the specifications of the hardware and software described in these manuals at any time and without prior notice. Nikon will not be held liable for any damages resulting from the use of this product. While every effort has been made to ensure that the information in these manuals is accurate and complete, we would appreciate it were you to bring any errors or omissions to the attention of the Nikon representative in your area (address provided separately).

Parts of the Camera

Power-on lamp ( Power switch ( Shutter-release button ( 16) Self-timer lamp ( 20) Built-in microphone ( 47) Eyelet for camera strap

OPTICAL 3X Z

9) 40, 47, 68)

Speaker (

:.2 m m 1

Built-in flash ( 18)

Lens ( 14, 85)

Lens cover closed

Battery-chamber cover ( Power cord channel cover for optional AC adapter ( 83)

Battery chamber ( 7)

Lens cover
Attaching the Camera Strap Attach the strap as shown below.
Replacing the connector cover
Monitor ( 4, 13, 17) Zoom button ( 14). MENU button See also: (thumbnail playback): ( 60) 44 (playback zoom): 45 (help): 62
Connector cover ( 48, 49, 54)

Flash lamp (

Multi selector ( 9) Memory card slot cover ( 12) (delete) button ( 17, 43, 44) 13, 23, 39) 23) ( (movie) mode 39)
Tripod socket ( (playback) button 17, 43) Mode selector ( (auto) mode 13)

(scene) mode ( (

The Multi Selector Press the multi selector up, down, left, or right to highlight pictures and menu items, then press the center to select. The buttons on the multi selector are also used to:
: Display self-timer menu ( 20) : Display macro closeup mode menu ( 22) ( : Display flash mode menu 18) : Perform D-lighting ( : Transfer picture ( 46) 49)

The Monitor

The following indicators may appear in the monitor during shooting and playback (actual display varies with current camera settings):

Shooting

Shooting mode. 13, 23, 39 Exposure lock.. 32 Macro close-up mode.. 22 Zoom indicator. 14, 22 Focus indicator.. 15 Flash mode. 18 Camera shake icon. 16 Internal memory/Memory card indicator.. 11 Battery level indicator.. 13 Sensitivity.. 16 Self-timer indicator. 20 Date not set indicator. 88 Time zone indicator. 64
13 Date imprint/Date counter. Number of exposures 21
remaining. 13, 98 Movie length. 39, 98 Focus area. 15 Exposure compensation. 75 Image mode.. 72 Color options.. 77 White balance.. 73 Best Shot Selector (BSS). 76 Continuous shooting mode.. 75

Playback

OK : START

10.10.2006 12:00

100NIKON 9999.JPG

9999 / 9999

D-Lighting icon. 46 Movie playback indicator. 40 Voice memo icon. 47 Image mode.. 72 Movie indicator. 40 Protect icon.. 80 Print-order icon. 57 Transfer icon.. 50, 80 Time of recording.. 9 Date of recording.. 9
1 Current folder. File number and type. Internal memory/Memory card 8 9
indicator.. 11 Battery level indicator.. 13 D-Lighting guide.. 46 Volume indicator.. 35, 40, 47 Voice memo recording guide. 47 Voice memo playback guide.. 47 Current frame number/Total number of frames/ Movie length.. 40

COPY Provides clear pictures of text or drawings on a white board or in print. Camera focuses on subject in center of frame; use macro close-up ( 22) to focus at short ranges. Colored text and drawings may not show up well. BACK LIGHT Use when light is coming from behind a portrait subject, throwing their features into shadow. The flash fires automatically to fill in (illuminate) shadows. Camera focuses on subject in center of frame.
PANORAMA ASSIST Use when taking a series of pictures that will later be joined side-by-side in PictureProject to form a panorama ( 32).
VOICE RECORDING Use to record, play back and copy audio only. The internal memory can hold approx. 50 minutes of voice recording, while a memory card (256MB or more) can hold up to approx. five hours.
* Focus indicator ( 15) is always displayed in green when shutter-release button is pressed halfway; note, however, that foreground objects may not always be in focus.
Taking Pictures for a Panorama ( 31) The camera focuses on the subject in the center of the frame. Use a tripod for better results.
Highlight (Panorama assist) and press the center of the multi selector.
Highlight the desired pan direction and press the center of the multi selector. The arrow showing the selected direction turns white.
The direction can be re-selected by pressing the center of the multi selector again. If the first shot is taken without deciding the pan direction, it will be automatically set for the right.
Take the first picture. The first picture is displayed in about 1/3 of the monitor. AE-L icon is displayed to show that white balance and exposure have been locked at values for first shot.
Compose the next shot to overlap with the previous picture and take the second picture. Repeat until the entire series of pictures has been taken.
End series. The monitor returns to the screen in step 1.
Panorama Assist Flash mode ( 18), self-timer ( 20) and macro close-up ( 22) cannot be adjusted after the first picture is taken. Pictures cannot be deleted, or zoom or image mode ( 72) cannot be adjusted until shooting is complete.
Making a Voice Recording (

VOICE RECORDING

2006. 10. 10 20:18 Max rec time 03:44:41 MENU Exit OK Confirm ? Help MENU Back REC
Highlight (Voice Recording) in the scene menu.
The monitor will show a icon and the length of the voice recording (hr:min.:sec.) that can be made.
Recording indicator File number Length of recording Progress indicator
VOICE RECORDING REC 003 2006. 10. 10 20:18 00:00:04/03:44:37 Index: 01 OK
Date of recording Time remaining Index number
Press the shutter-release button to start voice recording.
The monitor turns off after five seconds of commencing recording. Pressing the shutter-release button halfway, or pressing any of the buttons will turn on the monitor again for about five seconds. To pause recording, press the center of the multi selector. Press again to resume recording.

Press the shutter-release button to end recording. Recording will end automatically if the internal memory or memory card becomes full, or after five hours of recording.
Notes on Voice recording Do not touch the microphone during voice recording. Voice recording is not possible if the internal memory/memory card has less than 10 seconds of recording time available, or the battery charge is insufficient. It is recommended to use the AC adapter EH-65A (optional) for extended recording. Test recording beforehand, to ensure proper function, is advised.
Copying Voice Recordings to a Computer Voice recordings cannot be copied to a computer using the bundled software, PictureProject. Set the USB option in the SET-UP menu Interface to Mass Storage ( 49) and copy the files directly from the camera. Once copied to a computer, voice recordings can be played back with a player that supports WAV format files (such as QuickTime). Indexes added during recording work only when playing back on the camera. Adding an index An index can be added by pressing the multi selector up, down, left or right during recording. The camera can locate the index during playback and begin playback from this point. The start is assigned an index of one; other indices are assigned in ascending order, to a maximum of 98. The current index number is shown in the monitor.
VOICE RECORDING REC 003 2006. 10. 10 20:18 00:00:04/03:44:37 Index: 05 OK

Playing Voice Recordings

File number

SELECT SOUND FILE

2006. 10. 10 20:18 Max rec time 03:44:41 MENU Back REC
001 2006. 09. 01 11:2006. 10. 10 01:58 MENU Copy OK Confirm

Date of recording

Highlight (Voice Recording) in the scene menu and press the center of the multi selector. The available recording time is displayed.
Press the button. Currently saved voice recordings are listed.

AUDIO PLAYBACK

Control icons File number Playback time Index number Progress indicator
004 2006. 10. 21 20:30 00:00:10/00:00:48 Index: 02 OK Confirm
Date of recording Length of recording Volume Index markers
Highlight the desired voice recording and press the center of the multi selector. The voice recording playback screen is displayed and playback starts. Volume can be adjusted with the button during voice recording playback and the controls of the voice recording playback are displayed at the top of the monitor; press the multi selector left or right to highlight a control, and press the center of the multi selector to perform the selected operation.

Pause playback

Use multi selector to select icon, then press the center of the multi selector to pause playback.

Resume playback

When playback is paused, use multi selector to select icon, then press the center of the multi selector. Use multi selector to select icon, then press and hold the center of the multi selector. Use multi selector to select icon, then press and hold the center of the multi selector. If playback reaches the end, playback stops and voice recording list screen appears. Use multi selector to select icon, then press the center of the multi selector to return to start of current index. icon, then press Use multi selector to select the center of the multi selector to skip to next index. Use multi selector to select icon, then press the center of the multi selector to return to voice recording list.

Copy If there is not enough space in the memory or on the memory card to receive the copy, the message SOUND FILE CANNOT BE SAVED will be displayed. Delete unwanted pictures or voice recordings, or insert a new memory card (if using a memory card) before trying again. Copying of voice recordings taken with another make of camera is not guaranteed by Nikon.

Movies

Movie Mode
Follow the steps below to shoot movies with sound recorded via the built-in microphone. Movies recorded to internal memory at the default setting of Small Size 320 can be up to 16 seconds long; a 256MB memory card can hold up to 14 minutes 16 seconds.
Select mode. Exposure count display shows maximum total length of movie that can be recorded.

0h 1m22s

Start recording. Progress bar shows amount of memory remaining.

0h 0m54s

End recording. The monitor returns to the screen in step 1.
See Movie Options: The Movie Menu ( and focus options.
41) for information on movie size
Recording Movies The flash ( 18) and self-timer ( 20) are not available in movie mode. Macro closeup can be used ( 22). Optical zoom cannot be adjusted during recording. See the Appendix for information on the maximum total length of movie that can be recorded ( 98).

Movie Playback

In full-frame playback ( 43), movies ( 39) are indicated by a icon.
10.10.2006 15:30 100NIKON 0001.MOV MOV

OK START 0h0m9s

To play a movie, display it in full-frame playback and press the center of the multi selector. Playback controls appear at the top of the display; press the multi selector left or right to highlight a control, then press the center of the multi selector to perform the selected operation.

Control Description

0h 0m10s

Volume indicator

Rewind movie while center of multi selector is pressed. Advance movie while center of multi selector is pressed. Playback pause. Advance paused movie one frame. Rewind paused movie one frame. Resume playback. End movie and return to full-frame playback.
Volume is controlled by the zoom button. Press increase.

to lower volume,

Movie Options: The Movie Menu
The movie menu contains the options shown below. To display the movie menu:

Delete picture View playback menu Exit to full-frame playback Create cropped copy (see below) Exit to shooting mode Shutter-release button MENU

Creating a Cropped Copy

If a icon is displayed in the monitor during playback zoom, the portion of the picture currently visible in the monitor can be saved to a separate file. Pressing the shutter-release button displays the dialog shown at right; highlight Yes and press the center of the multi selector to create a new picture containing only the area visible in the monitor.
Save this image as displayed?
Depending on the size of the original and the zoom ratio at the time the copy was created, copies will be 2,816 2,112 (COOLPIX L2 only), 2,592 1,944 (COOLPIX L3 only), 2,048 1,536, 1,600 1,200, 1,280 960, 1,024 768, 640 480, 320 240, or pixels in size. Cropped pictures are stored in JPEG format (compression ratio 1:8) as separate pictures from the original pictures.
Cropped Copies Cropped copies may not display or transfer correctly when viewed on other models of Nikon camera. See the Appendix for more information on cropped copies ( 99).
Enhancing Contrast: D-Lighting
If a icon appears on a picture during full-frame playback ( 43), Dlighting can be used to create a copy with enhanced brightness and contrast, brightening dark and backlit subjects.
D-Lighting OK Cancel OK Confirm

10.10.2006 15:30

100NIKON 0005.JPG
Confirmation dialog is displayed.
Highlight OK and press the center of the multi selector. The enhanced brightness copy is made.
During playback, copies created with D-lighting are indicated by a icon.
D-Lighting Copies created with D-lighting may not display or transfer correctly when viewed on other models of Nikon camera. See the Appendix for more information on D-lighting copies ( 99).
Voice Memos: Recording and Playback
The built-in microphone ( 2) can be used to record voice memos for pictures marked with a icon in full-frame playback ( 43).
shutter-release Record voice memo: A voice memo will be recorded while the button shutter-release button is held down. During recording, the REC icon blinks. Recording ends after about 20 seconds or when the shutter-release button is released.
shutter-release Play voice memo/End playback: Press the shutter-release button button while displaying a picture with a voice memo (marked with ) in the full-frame to play voice memo. Playback ends when the voice memo ends or the shutter-release button is pressed again. Change volume: Zoom buttons control volume during playback. Press to lower volume, to increase. Delete voice memo: The confirmation screen will be displayed when is pressed. Press the multi selector up or down to highlight No, , or Yes, then press the center of the multi selector. No: picture and voice memo will not be deleted : voice memo only will be deleted Yes: picture and voice memo will be deleted

Macintosh: Drag the untitled camera volume (NO NAME) into the Trash.

Printing Pictures

Users of PictBridge-compatible printers can connect the camera directly to the printer and print pictures without using a computer. Take photographs Use Print set to select pictures for printing ( 59) Connect printer ( Print pictures one at a time ( 55) 54) Print current print order (DPOF printing; 57)
Print multiple pictures ( 57)

Disconnect USB cable

Other Ways to Print Pictures Pictures can also be printed by: Inserting a memory card in the printer: If the printer is equipped with a card slot, a memory card can be inserted in the printer and pictures printed directly from the card. If the printer supports DPOF, the camera Print set option can be used to select pictures for printing ( 59). See the printer manual for details. Taking memory cards to a digital print service: If the service supports DPOF, the camera Print set option can be used to select pictures for printing ( 59). Transferring pictures to a computer: Once transferred to a computer ( 49), pictures can be printed using imaging software. See the application and printer manuals for details.
Printing the Date of Recording on Pictures Two options are available for printing the date of recording on pictures: Date imprint ( 67) and Print set>Date ( 59).

Date imprint

Must be set before picture is taken Date permanently imprinted on picture

Print set>Date

Set after picture is taken Date stored separately
Date appears whenever picture is printed Date only appears when picture is printed on DPOF-compatible printer The date only appears once; Print set>Date cannot be used to print two dates on pictures with a date imprint.

Connecting the Printer

Turn the printer on.
Set the Interface>USB option in the setup menu to PTP ( 71). See Menus ( 60) for more information on menu operations.
Turn the camera on. A PictBridge start-up screen will be displayed. The camera will then enter full-frame playback with the PictBridge logo displayed in the monitor.

PictBridge

Printing Pictures One at a Time
PRINT MENU Start print Copies Paper size OK Confirm
Select a desired picture.
* To choose picture from thumbnail list, press ( ) and use multi selector to highlight desired picture, or press ( ) to exit to full-frame playback.

Display print options.

MENU Back OK Start print
Press multi selector up or down to highlight option, press center to select:

Start print

Start printing. To cancel before all copies have been printed, press center of multi selector. Camera returns to PictBridge playback display when printing is complete; repeat from Step 1 to print additional pictures. Press multi selector up or down to choose number of copies (up to 9). Press center of multi selector to return to previous menu.

Help For help on the highlighted item, press the ( ) button. Press to display the options for the highlighted item, ( ) to exit, or MENU to return to shooting or playback.
Basic Camera Setup: The Setup Menu
The setup menu contains the following options:
1/3 SET-UP Shooting menu Menus -Welcome screen -Date -Monitor settings MENU Exit OK Confirm ? Help
Menus Welcome screen Date Monitor settings Date imprint Sound settings Blur warning Auto off Format memory/ Format card Language Interface Reset all Battery Type Firmware version
Choose how menus are displayed. Select screen displayed when camera is turned on. Set clock and choose home and destination time zones. Adjust monitor brightness and choose display style. Imprint date or birthday counter on pictures. Adjust sound and volume settings. Turn blur warning on or off. Choose when monitor will turn off to save power. Format internal memory or memory card. Choose language for camera menus and messages. Adjust settings for connection to television or computer. Restore camera settings to default values. Choose type of battery inserted in camera. Display camera firmware version. 71 71
Default Settings See the Appendix for information on setup menu defaults (
Choose how menus are displayed.
SET-UP Shooting menu Menus Welcome screen Date Monitor settings MENU Exit OK Confirm ? 1/3 WELCOME IMAGE SELECT MENUS Text Icons

Welcome Screen

Choose the welcome screen displayed when the camera is turned on. To select a picture from internal memory or the memory card to use as a welcome screen, choose Select an image (the image used in the welcome screen is a copy of the selected image and does not change when the original is deleted or the memory card is removed).
WELCOME IMAGE SELECT Disable welcome Nikon Animation Select an image

COOLPIX

Set the camera clock and choose home and travel destination time zones.

Date Time zone

DATE 10.10.2006 15:30
Set camera clock to current date and time ( 9). Choose home ( ) or travel destination ( ) time zone ( 100); turn daylight saving time on or off.
Date Time zone MENU Exit OK Confirm --

To switch between home (

) and travel destination (

) time zones:

To switch between home ( ) and travel destination ( ) time zones: Highlight or and press the center of the multi selector. The clock is set to the time in the selected time zone. The current selection is indicated by a dot.
TIME ZONE 10.10.2006 15:30 London, Casablanca DaylightSaving OK London, Casablanca DaylightSaving MENU Exit OK Confirm
* To change the home or travel destination time zone, highlight or , then press the multi selector right. Choose a time zone and daylight saving time as described in Basic Setup ( 9).
The Clock Battery The clock battery charges when the main battery is installed or an AC adapter is connected, and can provide several days of backup power after about 10 hours of charging.

Monitor Settings

The following options are available:

Photo info Brightness

Choose information displayed in monitor. Choose from five settings for monitor brightness.
MONITOR SETTINGS -Photo info 3 Brightness
Photo Info The following display options are available:

Show info

Indicators are displayed in monitor ( 4).
Display Shooting mode Playback mode

Hide info

Indicators are not displayed in monitor.

Date Imprint

Imprint date information on pictures as they are taken.
Off Date Date and time Date counter
DATE IMPRINT Off Date Date and time Date counter MENU Exit OK Confirm
Time and date do not appear on pictures. Date or date and time are imprinted on pictures taken while this option is in effect. Time stamp shows number of days between date of shooting and selected date (see below).
At settings other than Off, the option selected is indicated by an icon in the monitor during shooting.
Date Counter Pictures taken while this option is in effect are imprinted with the number of days remaining until or the number of days elapsed since a specified date. Use it to track the growth of a child or count down the days until a birthday or wedding.
DATE COUNTER 9 2006.10.19 Number of days MENU Exit OK Confirm
Press multi selector right to view stored dates Press multi selector right to view display options

Stored dates

Up to three dates can be stored. To set, highlight, press multi selector to right, and enter date ( 9). To select, highlight and press center of multi selector.

STORED DATES

2006.10.19 2006.12.24 ---------------

Display options

Highlight Number of Days, Years and Days, or Yrs, Mnths, Days and press center of multi selector.
Sample date counter time stamps are shown below.
09/10.10.2006 09/10.10.2006
Future date (days remaining)

Past date (days elapsed)

Date Imprint The date is recorded in the order selected in the DATE menu ( 64). Date imprint cannot be used to print date information on existing pictures. See Creating a DPOF Print Order: Print Set ( 59) for a comparison of date imprint and Print set>Date. Date imprinted at an Image mode setting of TV screen(640) may be difficult to read. Choose PC screen(1024) or larger when using date imprint.

The current setting is shown by an icon in the display.

White Balance

Process pictures to produce natural colors under different types of light.
1/2 WHITE BALANCE Auto White bal. preset Daylight Incandescent Fluorescent MENU Exit OK Confirm
Auto White bal. preset Daylight Incandescent Fluorescent Cloudy
White balance automatically adjusted to suit lighting conditions. Best choice in most situations. Neutral-colored object used as reference to set white balance under unusual lighting conditions ( 74). White balance adjusted for direct sunlight. Use under incandescent lighting. Use under most types of fluorescent lighting. Use when taking pictures under overcast skies.

Use with the flash.

At settings other than Auto, the current setting is shown by an icon in the display.
Preset White Balance Preset white balance is used with mixed lighting or to compensate for light sources with a strong color cast (for example, to make pictures taken under a lamp with a red shade look as though they had been taken under white light).
Place white or gray reference object under lighting that will be used during shooting.
Highlight White bal.preset.

WHITE BAL PRESET

Cancel Measure MENU Exit OK Confirm

Camera zooms in.

Highlight Measure.
Frame the reference object and press the center of the multi selector. The new white balance value is set.
* To restore most recent value for preset white balance, highlight Cancel and press center of multi selector. Preset White Balance Preset white balance cannot be measured with the flash.

Exposure Compensation

Exposure compensation allows you to adjust the exposure setting suggested by the camera from between 2.0 EV (underexposure) and +2.0 EV (overexposure) in increments of 1/3 EV.
EXP. +/+0.-0.3 MENU Exit OK Confirm

Continuous

Capture a fleeting expression on a portrait subject, photograph a subject that is moving unpredictably, or capture motion in a series of pictures.

AUTO. Other modes can be

Image Mode ( 72) and Movie Options ( 41) The following table lists the number of pictures or maximum total movie length that can be stored in internal memory and on a 256MB memory card, together with the size of still pictures.
Image mode/ Movie Options

Internal memory (23MB)

256MB memory Print size card (printed at 300dpi) (cm/in.)
2203 3min. 39s 7min. 15s 7min. 15s 14min. 16s 47min. 8s 2418 / 97 2216.5 / / 97 2216.5 / / / / 21
High (2816) (COOLPIX L2) High (2592) (COOLPIX L3) Normal (2816) (COOLPIX L2) Normal (2592) (COOLPIX L3) Normal (2048) PC screen (1024) TV screen (640) TV movie 640 TV movie 640 Small size 320 Small size 320 Smaller size 160
213 21s 42s 42s 1min. 22s 4min. 33s
* All figures are approximate. Maximum total movie length and number of pictures that can be stored varies with make of memory card and, in the case of JPEG images, with the type of scene recorded. If memory remains for more than 9999 pictures, exposure count display will show 9999.
Restrictions on Camera Settings ( 72) Restrictions apply to the following settings in mode:

Flash mode

Flash turns off at Continuous settings other than Single or when BSS is on. Original mode is restored when Single is selected or BSS is turned off. If self-timer is on, only one shot will be taken when shutter is released, regardless of settings chosen for Continuous and BSS. Original Continuous and BSS settings are restored when picture is taken or self-timer is turned off. Selecting Continuous or Multi-shot 16 turns BSS off. Original BSS setting is not restored when Single is selected. Continuous is set to Single when BSS is on. Original setting is not restored when BSS is turned off. White balance is set to Auto and cannot be adjusted when Black-and-white, Sepia, or Cyanotype is selected for Color Options. Original white balance is restored when Standard color or Vivid color is selected.

Self timer

Continuous BSS White balance
Image File and Folder Names Pictures, movies, voice recording and voice memos are assigned file names with three parts: a four letter identifier, a four-digit file number assigned automatically by the camera in ascending order, and a three-letter extension (e.g., DSCN0001.JPG).

Unless otherwise stated, all figures are for a camera with fully-charged EN-MH1 batteries operated at an ambient temperature of 25C (77F).
MH-71 Battery Charger (Option; Supplied with Camera in Some Regions) Rated input Charging output Applicable batteries Discharge time Charging time Operating temperature Approximate dimensions Length of cord Approximate weight
AC 100240V, 0.110.065A, 50/60Hz DC 1.2V/2000mA Nikon EN-MH1 rechargeable Ni-MH batteries Approximately 5 hours Approximately 2.5 hours 535C (+4195F) 30mm/2.8 3.1 1.2in. (W H D) Approximately 1800mm (70.9in.) 110g (3.9oz.), excluding power cable
Specifications Nikon will not be held liable for any errors this manual may contain. The appearance of this product and its specifications are subject to change without notice. EN-MH1 NiMH Batteries The capacity of NiMH batteries will drop if they are charged repeatedly without first being fully discharged. The batteries will return to normal once fully discharged. Note that NiMH batteries gradually lose their charge when not in use; refresh the batteries immediately before use.

Symbols

(auto) mode, 1316 (playback) button, 17, 40, 4347 (movie) mode, 39 (delete) button, 17, 44, 45 (wait) icon, 88
Date, 911, 64, 67 printing, 59, 68 Date imprint, 67 Default settings, 95, 97 Delete, 80 Deleting pictures, 17, 44, 80 in full-frame playback, 17, 44 selected pictures, 80 Design rule for Camera File System, 96 Digital Print Order Format, 5759, 96 D-lighting, 46
Format card, 70 Format memory, 70 Framing, 14, 2428 assist, 2428 guides, 2428

Help, 62

A/V cable, 48 A/V-out connect, 48 Accessories, 83 AE-L, 32 Assist Modes, 2428 Auto off, 69 Auto-focus mode, 42
Image mode, 23, 72 Indoor photography, 29 Interface, 4855, 71 Internal memory, 45, 11, 70, 98 capacity, 98 formatting, 70
Backlit subjects, 18, 31 Batteries, 6, 71 Battery level indicator, 13, 88 Battery Type, 71 Best Shot Selector, 30, 76 Black-and-white, 77 Blur, 16, 69, 88, 92 Blur warning, 69
E-mail, 73 Error messages, 8890 Exif 2.2, 96 Exposure compensation, 75 Extension, 99

JPEG, 102

Landscape, 27, 30 Language, 9, 70 Language, choosing, 9 Lens, 2, 86
Face-priority AF, 2527 Fill flash, 18 firmware, 71 Firmware version, 71 Flash, 18, 95, 97 lamp, 3, 19 mode, 18, 97 Focus, 4, 15, 42 area, 4, 15 face-priority, 2527 indicator, 15 lock, 16 mode, 42 mode movie, 42 Format, 12, 70

Macro close-up, 22 Memory card, 45, 11, 83, 98 approved, 83 capacity of, 98 formatting, 70 Inserting and removing, 11 slot, 12 MENU button, 3, 60 Menus, 60, 64 Mode selector, 3
Close-up, 22, 30 Color options, 77, 99 Computer, 4852, 71 copying pictures to, Connector cover, 3 Continuous, 75 Copy (playback menu), 82, 100 Cropping pictures, 45 Cyanotype, 77
Monitor, 3, 45, 13, 66, 87 brightness, 66 indicators in, 45 Monitor settings, 66 MOVIE menu, 41 Movies, 3942 recording, 39, 41 viewing, 40 Multi selector, 3, 9, 61
(scene) mode, 2338 Back light ( ), 31 Beach/Snow ( ), 29 Close up ( ), 30 Copy ( ), 31 Dusk/Dawn ( ), 30 Fireworks show ( ), 30 Landscape ( ), 27 Museum ( ), 30 Night landscape ( ), 30 Night portrait ( ), 28 Panorama assist ( ), 31 Party/Indoor ( ), 29 Portrait ( ), 2527 Sports ( ), 28 Sunset ( ), 30 Self timer, 20, 99 Self-timer lamp, 2, 21 Sepia, 77 SET-UP menu, 60, 6371 SHOOTING MENU, 60, 7277 Shutter-release button, 2, 16 Side show, 79 Small pic., 81, 100 Smear, 87 Sound, 39, 68 Sound settings, 68 Strap, Camera, 2 Support information, 1
Transfer marking, 80 pictures for, 71, 80 Tripod, 3
USB, 71 cable, 49, 54 connection, 4952, 71 connector, 3
Video mode, 48, 71 Vivid Color, 77 Voice memo, 47 Voice recording, 33 copying, 37 playing, 35 Volume, 40
Night photography, 18, 28, 30 Noise reduction, 16 Number of exposures remaining, 13
Panorama, 3133 Paper size, 5657 PictBridge, 5359, 96 PictureProject, 1, 4952 Playback, 17, 4347 full-frame, 43 menu, 60, 7882 movies, 40 on TV, 48 thumbnail, 44 zoom, 45 PLAYBACK MENU, 60, 7882 Portraits, 2527, 28 Power switch, 2, 13 Power-on lamp, 2, 13 Print set, 59, 78 Print size, 56, 98 Protect, 80 Protected pictures, 80, 100
Welcome screen, 63 White balance, 7375, 99 preset, 7375
Zoom, 14 button, 3, 14 digital, 14 indicator, 4, 14 optical, 14 playback, 45
Television, 48 connecting to, 48 Thumbnail playback, 44 Time zone, 911, 64, 100 Transfer, 4951 mark, 50, 71, 80

Reset all, 71, 9598

No reproduction in any form of this manual, in whole or in part (except for brief quotation in critical articles or reviews), may be made without written authorization from NIKON CORPORATION.
Printed in Belgium YP6B01(11) 6MA14611--

Voice of the Customers: Mining Online Customer Reviews for Product Feature-based Ranking
Ramanathan Narayanan Alok Choudhary Electrical Engineering and Computer Science Department Northwestern University 2145 Sheridan road, Evanston, IL, 60208, USA Abstract
Increasingly large numbers of customers are choosing online shopping because of its convenience, reliability, and cost. As the number of products being sold online increases, it is becoming increasingly difcult for customers to make purchasing decisions based on only pictures and short product descriptions. On the other hand, customer reviews, particularly the text describing the features, comparisons and experiences of using a particular product provide a rich source of information to compare products and make purchasing decisions. Online retailers like Amazon.com1 allow customers to add reviews of products they have purchased. These reviews have become a diverse and reliable source to aid other customers. Traditionally, many customers have used expert rankings which rate limited a number of products. Existing automated ranking mechanisms typically rank products based on their overall quality. However, a product usually has multiple product features, each of which plays a different role. Different customers may be interested in different features of a product, and their preferences may vary accordingly. In this paper, we present a feature-based product ranking technique that mines thousands of customer reviews. We rst identify product features within a product category and analyze their frequencies and relative usage. For each feature, we identify subjective and comparative sentences in reviews. We then assign sentiment orientations to these sentences. By using the information obtained from customer reviews, we model the relationships among products by constructing a weighted and directed graph. We then mine this graph to determine the relative quality of products. Experiments on Digital Camera and Television reviews from real-world data on Amazon.com are presented to demonstrate the results of the proposed techniques.

Kunpeng Zhang

1 Introduction
The rapid proliferation of Internet connectivity has led to increasingly large volumes of electronic commerce. A study conducted by Forrester Research[1] predicted that e-commerce and retail sales in the US market during 2008 were expected to reach $204 billion, an increase of 17% over the previous year. More customers are turning towards online shopping because it is convenient, reliable, and fast. It has become extremely difcult for customers to make their purchasing decisions based only on images and (often biased) product descriptions provided by the seller. Online retailers aim to provide consumers a comprehensive shopping experience by allowing them to choose products based on their specic needs like price, manufacturer, and other attributes. They also allow customers to add reviews of products they have bought. Customer reviews of a product are generally considered more honest, unbiased and comprehensive than descriptions provided by the seller. Furthermore, reviews written by other customers describe the usage experience and perspective of (non-expert) customers with similar needs. A study by comScore and Kelsey group[2] showed that online customer reviews have signicant impact on prospective buyers. However, as the number of customer reviews available increases, it is almost impossible for a single user to read all reviews and comprehend them to make informed decisions. Therefore, mining these reviews to extract useful information efciently is an important and challenging problem. The abundance of customer reviews available has led to a number of scholars doing valuable and interesting research related to mining and summarizing customer reviews[3, 4, 5, 6, 7, 8]. There has also been considerable work on sentiment analysis of sentences in reviews, as well as sentiment orientation of a review as a whole[9]. There has been relatively little work on ranking products 1

1 http://www.amazon.com

negative sentiments about the product being reviewed. Comparative sentences contain information comparing competing products in terms of features, price, reliability etc. Fig. 1 depicts a typical customer review highlighting the different kinds of sentences. After developing techniques to identify such sentences, we build a weighted and directed product graph(feature-based) that captures the sentiments expressed by customers in reviews. We perform a ranking of products based on the graph. Particularly, the main contributions in this work are: NLP and dynamic programming techniques to identify subjective/comparative sentences(product feature-based) in reviews and determine their sentiment orientations. Sentence classication techniques to build a weighted and directed graph which reects the inherent quality of products in terms of their features. A ranking algorithm that uses this massive graph to produce a ranking list of products based on each considered product feature. That is, the end result of the algorithm is a ranking list that a potential customer can use to determine the best products based on customers importance of one or more product features. The reminder of this paper is organized as follows. Section 2 contains a summary of related work. We present our techniques in Section 3. Section 4 contains the details of datasets and experimental results followed by the conclusions in Section 5.

Figure 1: A Customer Review from Amazon.com. Product Features and Subjective/Comparative Sentences are Highlighted.
automatically based on customer reviews. Our work is based on the our proposed ranking scheme, where products are ranked based on their overall quality. A product feature is dened as an attribute of a product that is of interest to customers. Even though an overall ranking is an important measure, different product features are important to different customers based on their usage patterns and requirements. And different products may be designed and rank differently based on the feature of interest. For instance, a digital camera that is ranked highly overall may have less-than-stellar battery life. Thus, both overall ranking and more detailed product feature based ranking are important. In this paper, we propose an algorithm that uses customer review text, mines tens of thousands of reviews, and provides ranking of products based on product features. For each product category, product features are dened and extracted in the data preprocessing step. Note that for most products, there is a standard set of features which are considered important and normally are provided with product descriptions. We then label each sentence with the product features described in it. We then identify four different types of sentences in customer reviews that are useful in determining a products rank: positive subjective, negative subjective, positive comparative, negative comparative. Subjective sentences are those sentences in which the reviewer expresses positive or 2

2 Related Work

Our work is partly based on and closely related to opinion mining and sentence sentiment classication. Extensive research has been done on sentiment analysis of review text[9, 10] and subjectivity analysis (determining whether a sentences is subjective or objective[22]). Another related area is feature/topic-based sentiment analysis[23], in which opinions on particular attributes of a product are determined. Most of this work concentrate on nding the sentiment associated with a sentence (and in some cases, the entire review). There has also been some research on automatically extracting product features from review text[5]. Though there has been some work in review summarization[3], and assigning summary scores to products based on customer reviews[14], there has been relatively little work on ranking products using customer reviews. To the best of our knowledge, there has been no focused study on product feature based ranking using customer reviews. The most relevant work is [5], where

the authors introduce a unsupervised information extraction system which mines reviews to build a model of important product features, and incorporate reviewer sentiments to measure the relative quality of products. Our work differs from theirs in the following aspects: (1) We use a keyword matching strategy to identify and tag product features in sentences, (2) We use different strategies to assign sentiment orientation to sentences, and incorporate comparisons between different products in our model and (3) Based on sentence classication, we build a product graph for each product feature. By mining this graph, we are able to comprehensively rank products based on reviews they have received.
Product Ranking Methodology Identifying Product Features
Product features are attributes that provide added functionality to products. In the product description, manufacturers tend to highlight their product features from different (often contradicting) perspectives. The combination of features available in a product inuences the purchasing decision of a customer. Some consumers typically seek an optimized combination of features to satisfy their needs, while others may focus on a single feature. There has been some research on automatically identifying the different features of a product that customers are concerned about[5]. There has been earlier research on identifying product features and feature sentences, which is not the focus of this paper. We assume that product features associated with a product domain are given. We manually gathered product feature sets for two product cateogories: Digital Cameras and Television based on the ofcial consumer reports2. This is a one-time pre-processing overhead that can be done relatively easily by someone being familiar with a product domain. Table 1 shows the different product features and their major associated synonym sets. The use of synonyms is motivated by the fact that customers use different words/spelling variants to describe product features. To test the effectiveness of our feature-nding mechanism, we randomly picked 1000 review sentences from our review pool and manually labeled each sentence with the product features described. This small dataset was used to evaluate the performance of our feature-nding strategy. The precision and the recall of the keyword strategy for digital camera data were 0.853 and 0.807 respectively. Therefore, we are able to nd a major portion of feature sentences using this simple yet effective strategy.

2 http://www.consumerreports.org/
In our review dataset containing 1516001 sentences, we observed that around 16% of sentences describe one or more of these features. To tag each sentence with features, we use a simple strategy: if the sentence contains one of the words/phrases in the synonym set for a product feature, we mark it as describing the feature. Since we have dened these features manually, we describe them in greater detail. For Digital Cameras, ash is an important feature for indoor and low-light photography. Battery life is a sought-after feature that details the kind of batteries used. Focus talks about auto-focus or manual focus capabilities. Lens is a critical factor for professional photographers purchasing high-end cameras. The Optical feature encompasses digital zoom and optical zoom. LCD represents the digital display/screen that lets a user see how a photo will look like. Resolution refers to the sharpness, or detail, of a picture. Burst is used to describe the rapid re and continuous shooting capabilities. Memory determines the number of pictures that can be taken and Compression determines how le size of a photo is shrunk. For the Television segment, the Sound feature is useful for users interested in audio quality (some TVs come with an extra set of speakers to create surround sound). Reectivity/Anti-glare is important for the viewing experience. Size represents the size, height, weight of a television screen. Connections means the number and type of input ports available for hooking up devices to the television. The richness/quality of the images displayed are described in Picture quality. Users are also interested in the remote control device available with the television. Resolution refers to the number of pixels or lines displayed on the screen. Adjustment is the ability/mode that expands or compresses an image to ll the screen better. Picture-in-picture(PIP) feature allows a user to watch two channels at once. FilmMode/CineMotion improves the movie-watching experience, which may be important to some users.

3.2 Sentence Labeling

Customers express their opinions about products in multiple ways. We identify two kinds of sentences that are useful while ranking products: Subjective Sentences and Comparative Sentences. Before doing this task, we use MxTerminator[11] to split reviews into sentences because a typical customer review comprises of several sentences. We formally dene the different types of sentences below: Denition 3.1. Subjective Sentence(SS) A sentence expressing direct praise or deprecation about a product. Ex. This camera has excellent shutter speed. Denition 3.2. Comparative Sentence(CS) A sentence which indirectly express an opinion by performing a 3
Table 1: Keywords Representing 10 Most Important Product Features for Digital Camera and Television Domains Digital Camera TV resolution|pixel|megapixel connection|input|output| component video|composite video|HDMI lens|wide angle|normal range adjustment|stretch|zoom|expand|compress optical|zoom|optical zoom|digital zoom lm-mode|frame|theatrical|3:2|pull-down| motion compensation|CineMotion memory|megabytes|MB pip|picture-in-picture|dual-tuner|pop| picture-outside-picture|two-tuner burst|continuous|shutter|recovery|motion|sport resolution|1080p|1080i|720p battery|batteries|power screen|anti-glare|reectivity|burn-in|shiny| screensaver|pixel-shift focus|exposure|manual|iso picture|image|picture quality|image quality LCD|screen sound|sound quality|speaker|stereo|audio compression|compress|jpeg size|height|width|depth|weight|inch ash|light remoter|remote|gear|universal comparison between two products. Ex. I think the coolpix is better than the canon sd1200. Denition 3.3. Product Comparative Sentence(P CS) A PCS is a comparative sentence and contains at least one product name. Ex. This TV has much better sound quality when compared to the sony bravia. propose a more rened technique to nd comparative sentences specically related to product comparisons in our previous work[21]. We use a dynamic programming technique (longest common subsequence) to identify product-product comparison pairs in a comparative sentence. We use only comparative sentences which contain at least one product name which is different from the product the sentence is describing while building our ranking model. In [21], we have shown that we get a precision of 82% and a recall of 80% approximately.

Identifying Comparative Sentences
There has been some earlier research[16, 17] regarding identication of comparative sentences in text. These techniques use keyword comparison(KW contains 126 words, some of which are explicit(outperform, exceed, compare, superior, etc.) and others are implicit(prefer, choose, like, etc.)), sentence semantics, and sentence structure to identify comparative sentences. To identify part-of-speech tags, CRFTagger[12], a java-based conditional random eld part-of-speech (POS) tagger for English is employed. We build on these techniques and use the following rules for identifying comparative sentences: Check if the sentence contains any comparative keywords in KW ; Recognize any words with POS tags { JJR(comparative adjective), RBR(comparative adverb), JJS(superlative adjective), RBS(superlative adverb) }; Scan if any predened structural patterns are present in the sentence (as <word> as, the same as, similar to, etc.). Note that not all sentences satisfying these rules are comparative sentences in terms of product comparison. For example, the sentence I bought this camera for my son because he got a higher grade in his second statistical exam. does not show any comparative meanings or implications over other camera products. Therefore, we 4
3.4 Identifying Sentence Sentiment Orientation
In this section, we describe how we assign sentiment orientations for a sentence. We only consider positive and negative sentiments in this work. Unfortunately, dictionaries and other sources like WordNet[13] do not include sentiment orientation information for each word. Some researchers[18] have used supervised learning algorithms to infer the sentiment orientation of adjectives and adverbs from constraints on conjunctions. [9] contains a summary of existing sentiment analysis techniques. In this paper, we use a simple yet powerful method by utilizing a positive word set(POS) and a negative word set(NEG) developed in the MQPA project[19]. We also add some words of our own. At the end of this process, we get a list of 1974 words for the positive set and 4605 words for the negative set. We use a simple technique to identify the orientation of a sentence using these words. If the sentence contains a word that is in the positive word set, we label this sentence with a positive tag. Negative sentiment words are handled similarly. However, many customers do not like to express their opinions by using assertive sentences but using some negations in their reviews. In this case, the orientation should be switched. We constructed a set of 28 negation

words manually. It should be mentioned that we determine the sentence orientation for comparative sentences as well using the same list of sentiment words. A precision of 82% and a recall of 86% approximately can be achieved and shown in [21].
Table 2: Product Ranking Results for Gf in Fig. 2 Rank Vertex ID Score 1 B 0.D 0.C 0.A 0.052781 weight is also crucial to the ranking, in addition to the edge weights. The idea can be formalized using the equation below: pRank(P ) = [(1 d) + d pRank(Pi ) Ce (Pi )] Cv (P ), where
Constructing the Product Graph
We use the subjective and comparative sentences found to construct a directed and weighted graph for each product feature that can be mined to reveal the relative quality of products. The product graph for feature f is dened as follows: Gf = (V, E) where V is a set of nodes, V = {pi | each node represents a product, 0 < i < n}, E is a set of node pairs, called arcs or directed edges. An arc e = (pi , pj ) is considered to be directed from pi to pj. E = {ek = (pi , pj ), | Wei is the weight of the edge ei , 0 < i, j < n, 0 < k < m}, where n is the number of products, m is the number of edges. Consider a comparative sentence in the reviews for product Pi describing feature f. If this sentence compares product Pi with product Pj , we add a directed edge from Pj to Pi. The second step is to assign a weight to this edge. A comparative sentence occurring in the reviews for product Pi and comparing it with product Pj is considered a positive comparative(P C(Pi , Pj )) if it implies that Pi is better than Pj. If it implies that Pi is worse that Pj , it is considered a negative comparative(N C(Pi , Pj )). For each edge(Pj , Pi ), we count the number of positive (P C) and negative (N C) comparative sentences associated with the pair (Pi , Pj ) respectively. We assign the ratio P C/N C as the weight of the edge linking Pj to Pi. The weight of a node is used to represent the inherent quality of a product. For a node Pi , we use the ratio of the number of positive(P S) and negative(N S) subjective sentences (P S/N S) as its weight.

1{Pi ,P }

pRank(P ) is the product ranking of product P ; pRank(Pi ) is the product ranking of product Pi and n is the number of incoming links on product P ; 1{Pi ,P } is an indicator function, s.t. 1{Pi ,P } = if there is a link from Pi to P otherwise
W (P ,P Ce (Pi ) = Pm eWei(Pi),Pj ) , where m is the number j=1 of outbound links on product Pi , Pj are the nodes pointed to from Pi and We (Pi , Pj ) is the weight of the edge (Pi , Pj ). It is the edge weight contributor to the ranking of product P ; ) Cv (P ) = Pn Wv (P,P t ,Pt ). It is the node weight t=1 Wv (P contributor to the ranking of product P.

Let us illustrate the ranking process using a simple example. We have four products(A, B, C, D) which we wish to rank according to product feature f. The numbers of positive/negative, subjective/comparative sentences labeled with feature f are listed below. P Sf (A) = 1, P Sf (B) = 2, P Sf (C) = 3, P Sf (D) = 4 N Sf (A) = 3, P Cf (B, A) = 3, P Cf (B, C) = 7 P Cf (B, D) = 3, P Cf (A, C) = 2, N Cf (B, C) = 2 Based on these sentence statistics, we build a product graph Gf (see Fig. 2). Edge weights are determined by comparative sentences, and node weights are determined by subjective sentences. Since the reviews of product C have 7 positive comparative sentences mentioning product B (and feature F ), and 2 negative comparative sentences mentioning B (and feature f ), there is an edge from C to B with weight 3.5. It must be mentioned that to prevent edges with innite length (when the number of negative comparative sentences is 0), we set the minimum value of the denominator to 1 while 5

Ranking Products

We evaluate the relative importance of each product according to feature f by using the pRank algorithm described below. The classic PageRank algorithm[20] treats all edges equally and does not take node weights into account. A node has a higher importance if it is pointed to from relatively important nodes. In pRank, we not only consider the relative importance among products, but also take the importance/quality of the product itself into account. This means that the node
Algorithm 1 pRank Rank Products for Feature f Require: Product Feature(f ), Product Category(Cat). Ensure: The ranking list of products belonging to category Cat for the feature f.
1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14: 15: 16: 17: 18: 19: 20:
Figure 2: A Simple Ranking Example With a Product Graph Gf Having 4 Products Regarding a Specic Feature f (+ means positive and - means negative) computing edge weights. By using our algorithm, we get the ranking score for each node shown in the Table 2. The ranking order(the smaller, the product better) for this graph is B D C A. From the graph, we clearly see that A, C, D are worse than B because all of them have edges pointing to B. D has more positive subjective sentences than A, C and their comparative weights with B are approximately equal. C has a better ranking than A because (i) two sentences say A is better than C and (ii) reviews for A contain 1 positive/3 negative subjective sentences while reviews for C contain 3 positive subjective sentences. Algorithm 1 below summarizes our ranking methodology.
XML Data = Download(Cat); SENT = Get Sentence(XML Data); LSENT = Label(SENT, F ); {P S, N S, P C, N C} LSEN T ; for each sentence s {P C, N C} do Find all product comparison pairs {pi , pj } using dynamic programming; Pair {pi , pj } + P os or N eg ; end for Count P Spi , N Spi , P Cpi ,pj , N Cpi ,pj related to all products; for each product pi do for each product pj do if i == j then M atrix[i, i] = P Spi /N Spi ; else M atrix[i, j] = P Cpj ,pi /N Cpj ,pi ; end if end for end for Ranking List = Rank(M atrix[]); return Ranking List;

Experiment Results

In this section we evaluate the performance of our ranking algorithm. We conduct our experiments on customer reviews from two different product categories (Digital Camera and Televisions) from Amazon.com. Further details about the datasets and the APIs used to generate this data can be found at Amazon.com3 and BrowseNodes.com4. The Digital Camera dataset contains 83005 reviews (for 1350 products) and Television dataset contains 24495 reviews (for 760 products) collected by August, 2009. Table 3 and Table 4 show the relevant statistics for these two datasets: total number of sentences, frequency of occurrence of different product features, number of subjective and comparative sentences and their sentiment orientations. To evaluate our ranking algorithm, we rst perform product ranking based on the overall quality. To determine the overall rank of a product, we include all comparative and subjective sentences in our database while constructing
3 http://www.amazon.com/ 4 http://www.browsenodes.com/
the product graph. There is no ltering done for product features. We then mine this overall graph Goverall using the ranking algorithm described in Section 3.6. To evaluate the effectiveness of this ranking strategy, we compare our results with a ranking performed by domain experts. The results indicate that our product ranking strategy achieves signicant agreement with evaluations done by subject experts with several years of experience and insight in their respective elds. Approximately, an average overlapping probability of 62% could be achieved for different price bins for cameras and televisions. More details about this evaluation can be found in[21]. In this paper, we focus on the feature-specic ranking obtained by mining the individual product graphs generated for each product feature. Intuitively, the feature-specic ranking should not be dramatically different from the overall ranking. If we have chosen a relevant set of product features that customers are interested in, then the top-ranked products in these lists should not rank badly in the overall list. However, it is quite likely that there are signicant differences in the ranking order of these products, especially at the top. To clarify this intuition, we give the following example: If a product ranks in the top 5 products according to feature f (Ex. lens), then the probability that it ranks in the 6
Table 3: Breakdown of Subjective/Comparative Sentences(Digital Camera) Feature/Overall No. of Sentences No. of Subjective Sentences No. of Comparative Sentences Positive Negative Positive Negative Flash 514 Battery 533 Focus 720 Lens 437 Optical 338 Lcd 216 Resolution 227 Burst 189 Memory 143 Compression 29 Digital Camera 10890 Feature/Overall Sound Screen Size Connection Resolution Picture Quality Remoter Adjustment PIP Film-Mode TV Table 4: Breakdown of Subjective/Comparative Sentences(TV) No. of Sentences No. of Subjective Sentences No. of Comparative Sentences Positive Negative Positive Negative 9162

bottom 5 products overall should be very low. Similarly, if a product has high overall rank, then it should rank Table 5: The Relative Feature Fraction(RF Ff ) for Digital Camera and TV highly according to some features. Another aspect of our feature ranking methodology is the relative importance of different product features. A valid question that can be asked is: Which product features are customers looking for when making their choices? To answer this question we dene two metrics, Relative Feature Fraction(RF F ) and Importance of Feature(IF ). Digital Camera Flash Battery Focus Lens Optical LCD Resolution Burst Memory Compression RF Ff 18.18% 15.96% 15.93% 13.67% 10.77% 9.72% 5.63% 5.40% 4.06% 0.67% TV Sound Screen Size Connection Resolution Picture Quality Remoter Adjustment PIP Film-Mode RF Ff 24.76% 16.10% 12.87% 11.24% 10.98% 8.90% 8.13% 3.04% 2.15% 1.82%
Denition 4.1. Relative Feature Fraction: RF Ff = PNf *100%, where Nf is the number of sentences laf Nf Table 6: Importance of Feature(IFf ) for Digital Camera beled with feature f. and TV Denition 4.2. Importance of Feature: IFf = |XYf | IFf 100, where X = {top 10% of overall ranked Digital Camera Features IFf TV Features |X| Lens 79.9 Size 78.7 products}, and Yf = {top 10% of products according to Resolution 79.8 Film-Mode 72.3 feature f }. Optical 77.5 Picture Quality 70.7 Focus 76.3 Connection 69.1 76.3 PIP 69.1 Table 5 shows the RF F for different features in Dig- Memory Burst 75.2 Sound 67.5 ital Camera and Television categories respectively. We Lcd 74.1 Remoter 67.5 observe that customers are mentioning ash and sound Flash 72.9 Adjustment 64.3 (for Camera and TV) product features in their reviews Battery 71.6 Screen 61.2 most often. However, a higher RF F does not always in- Compression 68.4 Resolution 61.2 dicate greater importance of a feature. The importance of 7
Table 7: Top 10 Rated Digital Cameras for Each Feature and Overall Ranking Lens Resolution Optical Focus
PENTAX K100D SUPER KODAK EASYSHARE ONE CANON POWERSHOT SD500 CANON POWERSHOT SD990IS NIKON COOLPIX L3 NIKON COOLPIX 8400 NIKON COOLPIX L1 KODAK EASYSHARE C653 NIKON COOLPIX P6000 NIKON COOLPIX L4 CANON POWERSHOT SD200 NIKON COOLPIX S4 NIKON COOLPIX L3 NIKON COOLPIX S600 KODAK EASYSHARE C653 NIKON COOLPIX P90 NIKON COOLPIX L4 SONY ALPHA A700K NIKON COOLPIX P50 PENTAX OPTIO 60 CASIO EXILIM PRO KODAK EASYSHARE ONE CANON POWERSHOT SD550 CANON POWERSHOT SD990IS NIKON COOLPIX 3200 NIKON COOLPIX L3 NIKON COOLPIX S500 HP PHOTOSMART M537 NIKON COOLPIX P50 NIKON COOLPIX 7600 CASIO EXILIM PRO KODAK EASYSHARE ONE PANASONIC DMC-FX37A 10.1MP CANON POWERSHOT S60 CANON POWERSHOT SD990IS NIKON COOLPIX 3200 PENTAX OPTIO P70 NIKON COOLPIX L3 NIKON COOLPIX 8400 NIKON COOLPIX L1

Memory

KODAK EASYSHARE ONE CANON POWERSHOT S80 PANASONIC DMC-FX37A 10.1MP PENTAX OPTIO A10 CANON POWERSHOT SD990IS NIKON COOLPIX 3200 OLYMPUS SP-550UZ 7.1MP NIKON COOLPIX 4300 CANON POWERSHOT S410 NIKON COOLPIX 8400
CASIO EXILIM PRO CANON POWERSHOT SD500 PANASONIC DMC-FX37A 10.1MP CANON POWERSHOT S80 CANON POWERSHOT S60 CANON POWERSHOT SD990IS NIKON COOLPIX 3200 NIKON COOLPIX 995 CANON POWERSHOT S100 NIKON COOLPIX L3
KODAK EASYSHARE ONE PANASONIC DMC-FX37A 10.1MP CANON POWERSHOT S80 PENTAX OPTIO A10 CANON POWERSHOT SD990IS NIKON COOLPIX 3200 CANON POWERSHOT S400 CANON POWERSHOT G3 SONY DSCP150 7MP NIKON COOLPIX S500
KODAK EASYSHARE ONE CASIO EXILIM PRO PANASONIC DMC-FX37A 10.1MP NIKON COOLPIX S200 SONY DSCP150 7MP CANON EOS 1D NIKON COOLPIX 8400 NIKON COOLPIX L1 NIKON COOLPIX P90 SONY ALPHA A700K

Battery

CANON POWERSHOT SD990IS NIKON COOLPIX S4 NIKON COOLPIX S500 NIKON COOLPIX L3 KODAK EASYSHARE C653 HP PHOTOSMART M537 NIKON COOLPIX P90 NIKON COOLPIX P6000 NIKON COOLPIX L4 SONY ALPHA A700K

Compression

CANON POWERSHOT A620 CANON POWERSHOT SD300 KODAK EASYSHARE ZD710 CANON POWERSHOT S100 NIKON COOLPIX 8700 NIKON COOLPIX 4300 NIKON COOLPIX L3 NIKON COOLPIX P50 CANON POWERSHOT S230 CANON POWERSHOT A95

Overall Quality

NIKON COOLPIX L1 PANASONIC DMC-FX37A 10.1MP CANON POWERSHOT A990IS NIKON COOLPIX P90 HP PHOTOSMART M537 CANON POWERSHOT A70 NIKON COOLPIX 8800 FUJIFILM FINEPIX A330 KODAK EASYSHARE C653 OLYMPIS STYLUS 550
a feature should reect the role a feature plays in inuencing a customer. The IF metric measures the agreement between the overall ranking and feature-specic ranking. The values for IF for different product features are shown in Table 6. They indicate that our product feature-based ranking is consistent with overall ranking (which is veried previously by comparing with expert ranking [21]). In addition, this ratio should not be close to 1 because overall quality is different from feature quality. Table 6 also tells us that lens (Digital Camera) and size (Television) are the leading factors inuencing the overall quality of a product. Table 7 shows the top 10 cameras according to the overall ranking and featurespecic ranking. These results are consistent with the arguments made earlier.
a methodology and algorithm to rank products based on their features using customer reviews. First, we manually dene a set of product features that are of interest to the customers. We then identify subjective and comparative sentences in reviews using text mining techniques. Using these, we construct a feature-specic product graph that reects the relative quality of products. By mining this graph using a page-rank like algorithm(pRank), we are able to rank products. We implement our ranking methodology on two popular product categories (Digital Camera and Television) using customer reviews from Amazon.com. We believe our ranking methodology is useful for customers who are interested in specic product features, since it summarizes the opinions and experiences of thousands of customers.

Conclusion

6 Acknowledgement
This work was supported in part by NSF grants CNS0551639, HECURA CCF-0621443, OCI 0956311, SDCI OCI-0724599, OCI-0956311, IIS-0905205, and DOE grant DE-FG02-08ER25848. 8
Recent trends have indicated that large numbers of customers are switching to online shopping. Online customer reviews are an unbiased indicator of the quality of a product. However, it is difcult for users to read all reviews and perform a fair comparison. We describe

References

[1] Forrester research, http://www.comscore. com/Press_Events/Press_Releases/2007/ 11/Online_Consumer_Reviews_Impact_ Offline_Purchasing_Behavior. [2] comscore and Kelsey, http://www. shop.org/c/journal_articles/ view_article_content?groupId= 1&articleId702&version=1.0. [3] M. Hu and B. Liu, Mining and Summarizing Customer Reviews, Proceedings of the 10th ACM International Conference on Knowledge Discovery and Data Mining (SIGKDD-2004), 8 (2004), pp. 168174. [4] M. Hu and B. Liu, Mining Opinion Features in Customer Reviews, Proceedings of the 19th National Conference on Articial Intelligence., 7 (2004), pp. 755-760. [5] A. Popescu and O. Etzioni, Extracting product features and opinions from reviews, Proceedings of the conference on Human Language Technology and Empirical Methods in Natural Language Processing., (2005), pp. 339-346. [6] B. Liu, M. Hu, and J. Cheng, Opinion Observer: Analyzing and Comparing Opinions, WWW., 5 (2005), pp. 342351. [7] S. Kim, P. Pantel, T. Chklovski, and M. Pennacchiotti, Automatically Assessing Review Helpfulness, EMNLP., 7 (2006), pp. 423-430. [8] B. He, C. Macdonald, J. He, and I. Ounis, An Effective Statistical Approach to Blog Post Opinion Retrieval, CIKM., 10 (2008), pp. 1063-1069. [9] B. Liu, Sentiment Analysis and Subjectivity, to appear in Handbook of Natural Language Processing, Second Edition, 2010. [10] B. Pang, L. Lee, and S. Vaithyanathan, Thumbs up? Sentiment Classication using Machine Learning Techniques, EMNLP., (2002), pp. 79-86. [11] J. Reynar and A. Ratnaparkhi, A Maximum Entropy Approach to Identifying Sentence Boundaries., In Proceedings of the Fifth Conference on Applied Natural Language Processing , March 31-April 3, 1997. Washington, D.C. [12] X. Phan, CRFTagger: CRF English POS Tagger, http: //crftagger.sourceforge.net/, 2006. [13] G. A. Mille, C. Fellbaum, etc., WordNet: An Electronic Lexical Database, 5 (1998),MIT Press. [14] M. Hu, and B. Liu, Opinion extraction and summarization on the web, AAAI., 7 (2006), pp. 1621-1624. [15] B. Liu, X. Li, W. S. Lee, and P. S.Yu, Text Classication by Labeling Words, AAAI., 2004, pp. 425-430.

[16] N. Jindal, and B. Liu, Identifying Comparative Sentences in Text Documents, SIGIR., 8 (2006), pp. 244-251. [17] N. Jindal, and B. Liu, Mining Comparative Sentences and Relations, AAAI., (2006), pp. 1331-1336. [18] V. Hatzivassiloglou, and K. R. McKeown, Predicting the semantic orientation of adjectives, Proceedings of the eighth conference on European chapter of the Association for Computational Linguistics, 1997, pp. 174-181. [19] MPQA corpus http://www.cs.pitt.edu/mpqa, 2002. [20] L. Page, S. Brin, R. Motwani, and T. Winograd, The PageRank citation ranking: Bringing order to the Web, Stanford University, 1999. [21] K. Zhang, R. Narayanan, and A. Choudhary, Mining Online Customer Reviews for Ranking Products, Technical Report, EECS department, Northwestern University, 2009. [22] S. Arora, M. Joshi, and C. P. Rose, Identifying Types of Claims in Online Customer Reviews, NAACL HLT., 6 (2009), pp. 37-40. [23] Y. Lu, C. Zhai, and N. Sundaresan, Rated Aspect Summarization of Short Comments, WWW., 4 (2009), pp. 131140.