13 naacl-a latent variable model-qiu and jiang-slides
Download as PPTX, PDF1 like364 views
Threaded discussion forums provide an important social media platform. Its rich user generated content has served as an important source of public feedback. To automatically discover the viewpoints or stances on hot is-sues from forum threads is an important and useful task. In this paper, we propose a novel latent variable model for viewpoint discov-ery from threaded forum posts. Our model is a principled generative latent variable model which captures three important factors: view-point specific topic preference, user id and user interactions. Evaluation results show that our model clearly outperforms a number of baseline models in terms of both clustering posts based on viewpoints and clustering users with different viewpoints.
![Reference
? [Paul et al., AAAI?10] Paul, M. J. and Girju, R. (2010). A twodimensional topic-aspect model for discovering multi-faceted topics.
In AAAI.
? [Abu-Jbara et al., ACL?12] Amjad Abu-Jbara et al. (2012), Subgroup
detection in ideological discussions. In ACL.
? [Yi Fang et al. WSDM?12] Yi Fang et al. (2012), Mining contrastive
opinions on political texts using cross-perspective topic model. In
WSDM, pages 63¨C72.
? [Abu-Jbara et al., ACL?12] Amjad Abu-Jbara et al., (2012). Subgroup
detection in ideological discussions. In ACL.
? [Hassan et al., EMNLP?12] Hassan et al., (2012). Detecting
subgroups in online discussions by modeling positive and negative
relations among participants. In EMNLP.
25](https://image.slidesharecdn.com/13-naacl-alatentvariablemodel-qiuandjiang-slides-131107223649-phpapp01/85/13-naacl-a-latent-variable-model-qiu-and-jiang-slides-25-320.jpg)
More Related Content
Similar to 13 naacl-a latent variable model-qiu and jiang-slides (20)
Recently uploaded (20)
![[Webinar] Scaling Made Simple: Getting Started with No-Code Web Apps](https://cdn.slidesharecdn.com/ss_thumbnails/webinarscalingmadesimplegettingstartedwithno-codewebapps-mar52025-250305183437-f03c78a3-thumbnail.jpg?width=560&fit=bounds)
13 naacl-a latent variable model-qiu and jiang-slides
- 1. A Latent Variable Model for Viewpoint Discovery from Threaded Forum Posts Minghui Qiu and Jing Jiang School of Information System Singapore Management University 1
- 2. Threaded Forums ? Threaded structure ? With ?reply-to? relations (User interactions) ? Multiple threads on the same issue 2
- 3. Contrastive viewpoints in Threaded Forums Each Coin Has Two Sides the Chinese athlete Liu Xiang quit the London Olympic game Pro Obama or Anti Obama? How to find contrastive viewpoints from threaded forum posts? 3
- 4. Task and Method Overview Finding viewpoints for posts Finding viewpoints for users A set of corpus on one controversial issue Method ? A unified model for finding contrastive viewpoints (two-viewpoint) from threaded forum posts ? We build our model based on three observations 4
- 5. Observation 1: Different Viewpoints Will Have Different Topic Preference ? Our findings on ``LiuXiang¡± data set (``Will you support LiuXiang after he failed in London Olympic game???) 0.16 0.14 0.12 disappointed, athlete, ad sponsors Support LiuXiang Against LiuXiang Olympic hero, sympath y on his injury 0.1 0.08 0.06 0.04 0.02 0 21 34 39 28 22 6 19 31 4 37 14 8 16 12 13 30 17 11 7 18 Topic focus of two viewpoints on ¡°LiuXiang¡± Data Set 5
- 6. Observation 1: Different Viewpoints Will Have Different Topic Preference ? Framing1 ¨C Users with different sentiments/positions would focus on different aspects of the topic. E.g.: ¨C For ¡°iPhone¡± users: ¡°hardware and build¡±, ¡°siri¡±, ¡°ios¡± ¨C Against ¡°iPhone¡± users: ¡°physical keyboard¡±, ¡°android¡±, ¡°galaxy¡± ? Model assumption ¨C Each viewpoint has its own topic distribution 1D. Tversky, Amos; Kahneman. The framing of decisions and the psychology of choice. pages 453¨C458, 1981. 6
- 7. Observation 2: the Same User Will Hold the Same Viewpoint Towards an Issue ? User consistency ¨C Posts from the same user tend to have the same viewpoint towards an issue ¨C A viewpoint can be derived from the set of posts towards the same issue grouped by the same user ID ? Model assumption ¨C There is a user-level viewpoint distribution ¨C For each post by a user, its viewpoint is drawn from the corresponding user?s viewpoint distribution 7
- 8. Observation 3: User Interactions Reveal User Viewpoints ? User interaction ¨C User interaction: a post in reply to another user ¨C Users with the same viewpoint tend to have positive interactions among themselves, while with different viewpoint tend to have negative interactions ? Sample positive and negative interactions 8
- 9. Observation 3: User Interactions Reveal User Viewpoints ? Model assumption ¨C Interaction polarity is generated based on the viewpoint of the current post and the viewpoint of recipient post(s) User 1 Id 2 Viewpoint v1 User 2 Content Post Id V1 2 V1 V1 5 ? ¡ Positive Interaction 1 3 I agree with your post Dan. Obama is so ¡ Viewpoint ? p(POS): p(NEG): 1 - p(POS) Y 9
- 10. Overview of the Model ? A probabilistic model based on three observations ¨C Each viewpoint?s topic preference ¨C User consistency ¨C User interaction 10
- 11. Related Works ? Topic-Aspect Model (TAM, Paul et al., AAAI?10) ¨C A viewpoint-topic model where viewpoint and topic are orthogonal ¨C No user interaction ? Cross-Perspective Topic Model (Fang et al., WSDM?12) ¨C Supervised model ? Subgroup detection ¨C Mining user opinions (Abu-Jbara et al., ACL?12) ¨C User interaction (Hassan et al., EMNLP?12) ¨C Does not model viewpoints 11
- 12. A Probabilistic Model Topic specific word distribution Viewpoint specific topic distribution Y T w ?U: # of users ?N: # of posts ?L: # of words ?z: a topic label ?x: a switch ?x=0: w is background word ?x=1: w is topical word ?y: a viewpoint label ?s: a interaction type z x User-level viewpoint distribution L y s Interaction type N U The polarity of interaction type is learnt beforehand. 12
- 13. Polarity Prediction for Interaction Type ? Supervised learning ¨C Requiring labeled data ? Unsupervised approach ¨C Sample sentence: I agree with you ¨C Finding interaction expressions ? Finding sentences contains mentions of the recipient (user name or 2nd-person pronoun). E.g. you ? Surrounding words: a text window of 8 words. E.g.: I agree ¨C Interaction polarity ? Positive if there are more positive sentiment words, otherwise negative 13
- 14. Evaluation ? Data Sets ¨C English Data Sets ? Three most discussed threads from Abu-Jbara et al., ACL?12 ¨C Chinese Data Sets ? Three popular controversial issues in TianYaClub (one of the most popular Chinese online forums) ? Statistics 14
- 15. Data Annotation ? Identification of viewpoints ¨C 150 randomly sampled posts, two annotators (Cohen?s kappa agreement ¡Ý 0.61) ? Identification of user groups ¨C 150 randomly sampled users, two annotators (Cohen?s kappa agreement ¡Ý 0.70) To label a user?s viewpoint is easier than to label a post?s viewpoint 15
- 16. Baselines ? Topic-Aspect Model (TAM, Paul et al., AAAI?10) ¨C A viewpoint-topic model where viewpoint and topic are orthogonal ? Degenerate variants of our model ¨C UIM: User interaction model (part of our model) ¨C JVTM: Joint viewpoint-topic model (our model without interaction) ¨C JVTM-G: JVTM with a global viewpoint distribution 16
- 17. Identification of Viewpoints ? Task ¨C To identify each post?s viewpoint ? Results ? Our model significantly outperforms other models (at 10% significance level) ? Effectiveness of assumptions ? ? ? Each viewpoint¡¯s topic preference: JVTM > TAM User consistency: JVTM > JVTM-G User interaction: JVTM-UI > others ? User interaction is more important than other factors Averaged results of the models in identification of viewpoints 17
- 18. Identification of User Groups ? Subgroup detection ¨C To detect ideological subgroups, i.e.: user groups with different viewpoints ? Results ? Our model significantly outperforms other methods (at 10% significance level) ? Effectiveness of assumptions ? ? ? Each viewpoint¡¯s topic preference: JVTM > TAM User consistency: JVTM > JVTM-G User interaction: JVTM-UI > others Averaged results of the models in identification of viewpoints 18
- 19. Qualitative Analysis ? User interaction network on ¡°will you vote obama¡± Green (left) and white (right) nodes represent users with two different viewpoints discovered by our model. Red (thin) edges represent negative interactions while blue (thick) edges represent positive interactions More intra-cluster positive interactions and More inter-cluster negative interactions 19
- 20. Qualitative Analysis ? Users with different viewpoints tend to have different topic focus 0.16 Support LiuXiang 0.14 Against LiuXiang 0.12 0.1 0.08 0.06 0.04 0.02 0 21 34 39 28 22 6 19 31 4 37 14 8 16 12 13 30 17 11 7 18 Topic focus of two viewpoints on ¡°LiuXiang¡± Data Set 20
- 21. Qualitative Analysis ? Top 4 topics for ¡°supporting LiuXiang¡± viewpoint Word Translation Word Translation Word ÁõÏè LiuXiang À¸ hurdle Ô˶¯Ô± athlete µÚÒ» first ¹Ú¾ü champion ÉË injury °ÂÔË»á ʱ¼ä time Èüºó after-game ³É¼¨ record ¸úëì Olympic Achilles's tendon °ÂÔË Olympic Ìﾶ ˤµ¹ fall ±±¾© beijing »ñµÃ achieve ÄÐ×Ó track and field man 13Ãë 13s ½Å foot Ò»¸ö one ×îºó finally ÊÖÊõ surgery london ½ì time Áõ liu ¾öÈü final Â×¶Ø ÌïÁª IAAF Çé¿ö condition °ÂÔË»á Olympic Ó¢¹ú Britain Ò½Éú doctor train ²Î¼Ó attend ÊÜÉË hurt ÉϺ£ Shang Hai ѵÁ· ÖØ ÅÜ run field µ¼Ö result in already broken ¼ÇÕß ºÃ reporter ÒѾ Èü³¡ ¶ÏÁÑ good Òź¶ ¼Í¼ record Ó¢ÐÛ hero ÍÅ¶Ó team ÁªÈü 12Ãë 12s first heat ÐèÒª that time retire ¶á¹Ú Ìø ÅܵÀ champion µ±Ê± ÍËÒÛ Ô¤Èü 2012Äê ÂÞ²®Ë¹ pity league matches need jump report µÚ¶þ ΰ´ó 2nd great 2012 Robles Translation Word Translation heavy 21
- 22. Qualitative Analysis ? Top 4 topics for ¡°against LiuXiang¡± viewpoint Word Ìû ÉçÇø Translation post community Word ·¢×Ô Ëæʱ Translation orgin from anytime Word ÌìÑÄ Â¥Ö÷ Translation tianya poster Word ÌìÑÄ µÖÖÆ Translation tianya Resist Èȵã hot ÀÏ°å boss è sneak Æ×Ó lier Χ¹Û apathetic ÕþÐ CPPCC Âè F**K Ìå̳ sports ɵ±Æ fool °ï those Ë® spam ×î Ç® Ë®¾ü Ц Âî Ëï×Ó least money spam laugh scold foolish medal Ψ½ðÅÆÂÛ gold theory only smile ΢Ц support ¶¥ nausea ¶ñÐÄ ¿É¿Ú¿ÉÀÖ Coca Cola drink ºÈ joke Ц»° Ëï×Ó Æ¡¾Æ Ñî È«¼Ò ±ðÓÐÓÃÐÄ ¶ã foolish bear yang whole family ulterior motive hide Ìá ³Ô ÅÆ ¿àЦ ¸ßÉÐ ÓÐÁ¦ ÄãÃÇ you ¼ÓÓÍ cheer up Íá·ç bad tendency ÀÍÃñÉË²Æ ¶àô extremly ÍÑÀë ÓÐÈË someone ǹÑÛ Á³ÉÏ face Éñλ separate ¿´¿´ force of public ̲ opinion fame ¾«Éñ look ºÚ mention eat medal bitter smile noble powerful a waste of money and manpower spam those »Æ¼Ì¹â a hero spirit ÉñÏñ fame 22
- 23. Summary ? Conclusion ? A viewpoint discovery model for threaded forums ? Modeling three observations ? Viewpoint-specific topic distribution (Framing) ¨C User consistency ¨C Interplay between user interactions and viewpoints ¨C Future work ¨C ¨C ¨C ¨C Document representation: complex lexical units A more accurate interaction polarity classifier Contrastive viewpoint summarization Mining controversial issues and finding viewpoints 23
- 24. Thank you 24
- 25. Reference ? [Paul et al., AAAI?10] Paul, M. J. and Girju, R. (2010). A twodimensional topic-aspect model for discovering multi-faceted topics. In AAAI. ? [Abu-Jbara et al., ACL?12] Amjad Abu-Jbara et al. (2012), Subgroup detection in ideological discussions. In ACL. ? [Yi Fang et al. WSDM?12] Yi Fang et al. (2012), Mining contrastive opinions on political texts using cross-perspective topic model. In WSDM, pages 63¨C72. ? [Abu-Jbara et al., ACL?12] Amjad Abu-Jbara et al., (2012). Subgroup detection in ideological discussions. In ACL. ? [Hassan et al., EMNLP?12] Hassan et al., (2012). Detecting subgroups in online discussions by modeling positive and negative relations among participants. In EMNLP. 25
Editor's Notes
- #9: Users with the same viewpoint tend to have positive interactions among themselvesUsers with different viewpoints tend to have negative interactions among themselves
- #10: The polarity of an interaction expression is generated based on the viewpoint of the current post and the viewpoint of the post(s) that the current post replies to