0
Identifying        Consumers’ Arguments in Text               Jodi Schneider1 and Adam Wyner21 - Digital Enterprise Resear...
Outline• Motivation & Goals• Our Approach      – Provide a Semi-Automated Support Tool      – Use Argumentation Schemes   ...
Reviews are rich & detailedOctober 9, 2012          Schneider & Wyner, SWAIE at EKAW 2012   3
Customers disagree,                  especially in commentsOctober 9, 2012       Schneider & Wyner, SWAIE at EKAW 2012   4
Customer Questions• What’s controversial?• What are some reasons to buy the item? Not to buy it?• What sorts of people par...
Manufacturer Questions• What features are controversial?• What market segments report positive  (negative) experiences?• W...
Limited StructureOctober 9, 2012     Schneider & Wyner, SWAIE at EKAW 2012   7
Goal: A knowledge base we can query• Who likes this camera?• What statements are made about particular  camera features?  ...
Our approach• Build a support tool    – semi-automated    – rule-based    – using text analytics• Use argumentation scheme...
Simple Reasoning PatternPremises:• The Canon SX220 has good video quality.• Good video quality promotes image quality for ...
Argumentation SchemePremises:• The <camera> has <feature>.• <feature> promotes <user value> for <user class>.Conclusion:• ...
Variables as Targets for Information                        Extraction<camera><property><user value><user type><e-commerce...
4 Argumentation Schemes in the Paper1.     User Classification2.     Camera Classification3.     Appropriateness4.     Con...
Building more complex reasoning patterns   • “Cascade” of argumentation schemes   • Conclusions of one scheme as premises ...
Consumer Relativised                   Argumentation Scheme   3 Premises:          1. User Class (Conclusion of User Class...
Consumer Relativised                  Argumentation Scheme   Premises:   1. Cameras of class Y are appropriate for agents ...
Appropriateness Argumentation SchemeOctober 9, 2012   Schneider & Wyner, SWAIE at EKAW 2012   17
Appropriateness Argumentation Scheme   Premises:   1. Agent x is in user class X.   2. Camera y is in camera class Y.   3....
Premises become                  Information Extraction TargetsPremises of the Appropriateness AS:1. Agent x is in user cl...
Information Extraction   1.      User class   2.      (Camera class)   3.      Contexts of use: camera’s, user’s   4.     ...
Query for patternsOctober 9, 2012     Schneider & Wyner, SWAIE at EKAW 2012   21
Amazing low light photosOctober 9, 2012         Schneider & Wyner, SWAIE at EKAW 2012   22
Mainly bright colours in good daylightOctober 9, 2012   Schneider & Wyner, SWAIE at EKAW 2012   23
Arguments are User Relative• Amazing low light photos?• Only for bright colours in good daylight?•  Motivates the user cl...
Future work: argumentation schemes• Further instantiate the schemes using the tool      – Where do they work well?      – ...
Future work: ontologies & concepts• Ontologies and reasoning      – Ontology for users      – Ontology for cameras      – ...
Future work: evaluation• Evaluate the tool      – How well does it support users? (faster, better analyses?)      – Do ann...
Related Papers• Talk at EKAW, Thursday 11:45: “Dimensions of  argumentation in social media”  Schneider, Davis, and Wyner ...
Acknowledgements •     FP7-ICT-2009-4 Programme, IMPACT Project, Grant       Agreement Number 247228. •     Science Founda...
Thanks for your attention!• Questions?• Contacts:      – Jodi Schneider               jodi.schneider@deri.org      – Adam ...
October 9, 2012   Schneider & Wyner, SWAIE at EKAW 2012   31
4 Argumentation Schemes in the Paper1. User Classification AS2. Camera Classification AS3. Appropriateness AS      Conclud...
Domain terminologyOctober 9, 2012      Schneider & Wyner, SWAIE at EKAW 2012   33
Find camera features• Use                   :   – Has a flash   – Number of megapixels   – Scope of the zoom   – Lens size...
Find argument passages  after, as, because, for, since, when, ....• C  therefore, in conclusion, consequently, ....October...
Argument indicators:                  Premise & ConclusionOctober 9, 2012      Schneider & Wyner, SWAIE at EKAW 2012   36
To find attacks between arguments• Use contrast terminology:   – Indicators     but, except, not, never, no, ....   – Cont...
Sentiment terminologyOctober 9, 2012       Schneider & Wyner, SWAIE at EKAW 2012   38
,                                                          ,October 9, 2012   Schneider & Wyner, SWAIE at EKAW 2012       39
User Classification argumentation scheme   Variables are our targets for extraction.   Premises:      Agent x…   1.        ...
An argument for buying the cameraPremises:   The pictures are perfectly exposed.   The pictures are well-focused.   No cam...
An argument for NOT buying the                    cameraPremises:   The colour is poor when using the flash.   The images ...
Counterarguments to the premises of             “Don’t buy”             The colour is poor when using the flash.          ...
Making sense of reviews• Do other reviews agree?    – Any counterarguments?• Is this point relevant to me?    – Does this ...
Upcoming SlideShare
Loading in...5
×

Identifying consumers’ arguments in text swaie at ekaw 2012 10-09

533

Published on

A talk for SWAIE2012 at EKAW2012 http://semanticweb.cs.vu.nl/swaie2012/

Paper at http://jodischneider.com/pubs/swaie2012.pdf

Published in: Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
533
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
5
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide
  • Tuesday, October 9, 201230 mins for presentation including questions.http://jodischneider.com/pubs/swaie2012.pdfSWAIE: http://semanticweb.cs.vu.nl/swaie2012/
  • Why is opinion or sentiment analysis **not** sufficient? Because:It provides no explanation or justification for the opinion, broadly construed.We can count the numbers of participants who hold an opinion, but one well-made &apos;counter-argument&apos; may lead individuals to retract their opinion.Knowledge in the text is implicitly structured and many-layered. How can we extract that structured information?
  • AZW – I like having questions up front. However, to manage expectations, we don&apos;t want to ask questions we are not really addressing or questions that introduce complex issues. For instance, only derivatively do we inquire about &apos;who should i believe&apos; and &apos;why&apos;. It is derivative in the sense that this might be what people think about, but it is **not** in evidence in the surface of the data nor in the extractions we work with. We have, in this paper, nothing to say on this matter. How about:- What are some reasons to buy the item?What are some reasons not to buy the item?What sorts of people participate in the discussion?Are there authorities who can help me decide what to buy?Are there people who are similar to me who like/dislike this item and why?What are the opinions about features of the item?
  • From the manufacturer’s side, there is a related problem since she wishes tosell a product to a consumer. Looking at the reviews, the manufacturer must also extractinformation about specific topics from the corpus and structure the information into aweb of claims and counterclaims. With this information, the manufacturer could havefeedback about the features that the consumer does or doesn’t like, the problems thatthe consumer experiences, as well as the proposed solutions.
  • Replies are the main structure (tree-like)***Later: List of review attributes for Amazon reviews
  • We use 4 argumentation schemesUser ClassificationCamera ClassificationAppropriatenessCamera Relativised
  • Successively unpacking assumptions, arguments
  • Usedto tie the consumer&apos;s interests/properties to the camera&apos;s propertiesSimilarly we have a user relativised scheme, which uses this + user classification + camera classification to relativise the consumer to the camera.
  • Usedto tie the consumer&apos;s interests/properties to the camera&apos;s propertiesSimilarly we have a user relativised scheme, which uses this + user classification + camera classification to relativise the consumer to the camera.
  • Haven’t looked at camera class – corpus is 99 reviews for a single camera.
  • We use 4 argumentation schemesUser ClassificationCamera ClassificationAppropriatenessCamera Relativised
  • binary values (such as has a flash), properties with ranges (such as the number of megapixels, scope of the zoom, or lens size), and multi-slotted properties (e.g. the warranty).
  • Screenshot from GATE, in which we have built components of a toolPurple: conclusionOrange: premiseLots of ambiguity – different meanings of the words*DOES* draw attention to relevant places. Can turn on &amp; off particular things that we’re looking for. Helps with the search problem.
  • Drawn from vast lists of terminology, given sentiment valence: positive vs. negative +5 to 0 to -5Can look for various levels or homogenize – this is homogenized
  • We have an argument for buying the camera, an argument for not buying the camera. They rebut each other.We have attacks on the premises for “don’t buy the camera”. The argument for not buying the camera is defeated; the argument for buying the camera stands. So you should buy the camera.
  • Transcript of "Identifying consumers’ arguments in text swaie at ekaw 2012 10-09"

    1. 1. Identifying Consumers’ Arguments in Text Jodi Schneider1 and Adam Wyner21 - Digital Enterprise Research Institute, National University of Ireland, Galway 2 – Department of Computer Science, University of Liverpool Tuesday October 9, 2012 SWAIE 2012 (colocated with EKAW 2012) at National University of Ireland Galway, Ireland
    2. 2. Outline• Motivation & Goals• Our Approach – Provide a Semi-Automated Support Tool – Use Argumentation Schemes – Use Information Extraction• Example ResultsOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 2
    3. 3. Reviews are rich & detailedOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 3
    4. 4. Customers disagree, especially in commentsOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 4
    5. 5. Customer Questions• What’s controversial?• What are some reasons to buy the item? Not to buy it?• What sorts of people participate in the discussion?• Are there authorities who can help me decide what to buy?• Are there people similar to me who like this item? And why? …Similar people who dislike it? Why?• What opinions are given about features of the item?October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 5
    6. 6. Manufacturer Questions• What features are controversial?• What market segments report positive (negative) experiences?• What else are customers talking about? May reveal other customer needs. – Advice – Competitor’s products – Related products to be used in conjunction? October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 6
    7. 7. Limited StructureOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 7
    8. 8. Goal: A knowledge base we can query• Who likes this camera?• What statements are made about particular camera features? e.g. indoor picture quality• Which claims do they support? e.g. Do they support the claim that “the camera gives quality indoor pictures”? Or the opposite claim? October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 8
    9. 9. Our approach• Build a support tool – semi-automated – rule-based – using text analytics• Use argumentation schemes – patterns for reasoning – identify text mining targets for info extraction October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 9
    10. 10. Simple Reasoning PatternPremises:• The Canon SX220 has good video quality.• Good video quality promotes image quality for casual photographers.Conclusion:• Casual photographers should buy the Canon SX220.October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 10
    11. 11. Argumentation SchemePremises:• The <camera> has <feature>.• <feature> promotes <user value> for <user class>.Conclusion:• <user class> should <e-commerce action> the <camera>.<e-commerce action>: buy, not buy, sell, return, …October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 11
    12. 12. Variables as Targets for Information Extraction<camera><property><user value><user type><e-commerce action>October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 12
    13. 13. 4 Argumentation Schemes in the Paper1. User Classification2. Camera Classification3. Appropriateness4. Consumer RelativisedOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 13
    14. 14. Building more complex reasoning patterns • “Cascade” of argumentation schemes • Conclusions of one scheme as premises for anotherOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 14
    15. 15. Consumer Relativised Argumentation Scheme 3 Premises: 1. User Class (Conclusion of User Classification AS) 2. Camera Class (Conclusion of Camera Classification AS) 3. Appropriateness (Conclusion of Appropriateness AS) Conclusion: User should buy CameraOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 15
    16. 16. Consumer Relativised Argumentation Scheme Premises: 1. Cameras of class Y are appropriate for agents of class X. 2. Camera y is of class Y. 3. Agent x is of class X. Conclusion: Agent x should buy camera y.October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 16
    17. 17. Appropriateness Argumentation SchemeOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 17
    18. 18. Appropriateness Argumentation Scheme Premises: 1. Agent x is in user class X. 2. Camera y is in camera class Y. 3. The camera’s contexts of use satisfy the user’s context of use. 4. The camera’s available features satisfy the user’s desirable features. 5. The camera’s quality expectations satisfy the user’s quality expectations. Conclusion: Cameras of class Y are appropriate for agents of class X.October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 18
    19. 19. Premises become Information Extraction TargetsPremises of the Appropriateness AS:1. Agent x is in user class X.2. Camera y is in camera class Y.3. The camera’s contexts of use satisfy the user’s context of use.4. The camera’s available features satisfy the user’s desirable features.5. The camera’s quality expectations satisfy the user’s quality expectationsOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 19
    20. 20. Information Extraction 1. User class 2. (Camera class) 3. Contexts of use: camera’s, user’s 4. Features: camera’s available, user’s desirable 5. Quality expectations: camera’s, user’sOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 20
    21. 21. Query for patternsOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 21
    22. 22. Amazing low light photosOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 22
    23. 23. Mainly bright colours in good daylightOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 23
    24. 24. Arguments are User Relative• Amazing low light photos?• Only for bright colours in good daylight?•  Motivates the user classificationOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 24
    25. 25. Future work: argumentation schemes• Further instantiate the schemes using the tool – Where do they work well? – Improvements needed?• Develop additional schemes – Expertise – Comparison – Particular features (e.g. warranties)October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 25
    26. 26. Future work: ontologies & concepts• Ontologies and reasoning – Ontology for users – Ontology for cameras – Test inferences by importing scheme instances into an argumentation inference engine.• Address conceptual issues – Clarify distinctions between the camera’s quality expectations and features – Support matches between a user’s values and camera propertiesOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 26
    27. 27. Future work: evaluation• Evaluate the tool – How well does it support users? (faster, better analyses?) – Do annotation types match users’ expectations? (interannotator agreement)October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 27
    28. 28. Related Papers• Talk at EKAW, Thursday 11:45: “Dimensions of argumentation in social media” Schneider, Davis, and Wyner (EKAW 2012).• Wyner, Schneider, Atkinson, and Bench-Capon. “Semi-Automated Argumentative Analysis of Online Product Reviews.” In 4th International Conference on Computational Models of Argument (COMMA 2012).• Wyner and Schneider (2012). Arguing from a point of view, Agreement Technologies.October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 28
    29. 29. Acknowledgements • FP7-ICT-2009-4 Programme, IMPACT Project, Grant Agreement Number 247228. • Science Foundation Ireland Grant No. SFI/08/CE/I1380 (Líon- 2) • Short-term Scientific Mission grant from COST Action IC0801 on Agreement TechnologiesOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 29
    30. 30. Thanks for your attention!• Questions?• Contacts: – Jodi Schneider jodi.schneider@deri.org – Adam Wyner adam@wyner.infoOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 30
    31. 31. October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 31
    32. 32. 4 Argumentation Schemes in the Paper1. User Classification AS2. Camera Classification AS3. Appropriateness AS Concludes: Camera Class is appropriate for User Class Premises: User Class, Camera Class, User & Camera Match • Match on: Contexts of Use, Features, Quality Expectations4. Consumer Relativised AS Concludes: User should buy Camera Premises: User Class, Camera Class, AppropriatenessOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 32
    33. 33. Domain terminologyOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 33
    34. 34. Find camera features• Use : – Has a flash – Number of megapixels – Scope of the zoom – Lens size – The warrantyOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 34
    35. 35. Find argument passages after, as, because, for, since, when, ....• C therefore, in conclusion, consequently, ....October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 35
    36. 36. Argument indicators: Premise & ConclusionOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 36
    37. 37. To find attacks between arguments• Use contrast terminology: – Indicators but, except, not, never, no, .... – Contrasting sentiment The flash worked . The flash worked .October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 37
    38. 38. Sentiment terminologyOctober 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 38
    39. 39. , ,October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 39
    40. 40. User Classification argumentation scheme Variables are our targets for extraction. Premises: Agent x… 1. … has user’s attributes aP1; aP2; … 2. … user’s context of use aU1; aU2; … 3. … has user’s desirable camera features aF1; aF2; ... 4. … has user’s quality expectations aQ1; aQ2; ... 5. … has user’s values aV1; aV2; ... 6. …has desirable camera features aF1; aF2; … promote/demote user’s values aV1; aV2; ... Conclusion: Agent x is in class X.October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 40
    41. 41. An argument for buying the cameraPremises: The pictures are perfectly exposed. The pictures are well-focused. No camera shake. Good video quality. Each of these properties promotes image quality.Conclusion: (You, the reader,) should buy the CanonSX220.October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 41
    42. 42. An argument for NOT buying the cameraPremises: The colour is poor when using the flash. The images are not crisp when using the flash. The flash causes a shadow. Each of these properties demotes image quality.Conclusion: (You, the reader,) should not buy the CanonSX220.October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 42
    43. 43. Counterarguments to the premises of “Don’t buy” The colour is poor when using the flash. For good colour, use the colour setting, not the flash. The images are not crisp when using the flash. No need to use flash even in low light. The flash causes a shadow. There is a corrective video about the flash shadow.October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 43
    44. 44. Making sense of reviews• Do other reviews agree? – Any counterarguments?• Is this point relevant to me? – Does this reviewer have similar needs? – Does it apply in my situation?• Is enough information provided? – Any explanations? – Any examples? October 9, 2012 Schneider & Wyner, SWAIE at EKAW 2012 44
    1. A particular slide catching your eye?

      Clipping is a handy way to collect important slides you want to go back to later.

    ×