A MULTI-OBJECTIVE OPTIMIZATION APPROACH FORQUESTION ROUTING IN COMMUNITY QUESTIONANSWERING SEVICES
ABSTRACT
Community Question Answering (CQA) has increasingly become an important service for people asking questionsand providing answers online, which enables people to help each other by sharing knowledge. Recently, with accumulation ofusers and contents, much concern has arisen over the efficiency and answer quality of CQA services. To address this problem,question routing has been proposed which aims at routing new questions to suitable answerers, who have both high possibilityand high ability to answer the questions. In this paper, we formulate question routing as a multi-objective ranking problem, andpresent a multi-objective learning-to-rank approach for question routing (MLQR), which can simultaneously optimize the answeringpossibility and answer quality of routed users. In MLQR, realizing that questions are relatively short and usually attached with tags,we first propose a tagword topic model (TTM) to derive topical representations of questions. Based on TTM, we then developfeatures for each question-user pair, which are captured at both platform level and thread level. In particular, the platform-levelfeatures summarize the information of a user from his/her history posts in the CQA platform, while the thread-level features modelthe pairwise competitions of a user with others in his/her answered threads. Finally, we extend a state-of-the-art learning-to-rankalgorithm for training a multi-objective ranking model. Extensive experimental results on real-world datasets show that our MLQRcan outperform state-of-the-art methods in terms of both answering possibility and answer quality.
EXISTING SYSTEM:
Due to the importance of question routing in CQAservices, many approaches have been proposed in theliteratures. Existing studies on question routing can bebroadly categorized into two classes. The first classof studies mainly focuses on finding the users whoare more likely to respond to newly posted questions.Zhou et al. tackle question routing in a way ofclassification and capture question-user relationshipfeatures to decide whether a user will answer a question.Ji and Wang propose a learning to rankframework for question routing, and adopt pointwiseand pairwise approaches to learn the ranking models,which aims to find the potential answerers by takingasker-answerer pairs to conduct training instances. Liet al. design an incorporation of question categoryfor routing questions to potential answerers.Question recommendation is very similar to questionrouting, which can be also used to route questions tousers. The difference is that question recommendationconsiders CQA as a recommendation system, whereitems to be recommended to users are simply questions.Most of the works on question recommendationin CQA services are proposed to recommend questionsto the users who have high answering possibility.Dror et al. derive multiple features from themulti-channel attributes of the users and questions,then use these features to train a classifier to predictwhich users will answer the given question. To satisfydifferent types of answerers in CQA services, Szpektoret al.There are also a bunch of studies on expert find-ing in CQA services. Bouguessa et al. make acomparison on different link analysis methods forexpert finding, and adopt InDegree (i.e., the numberof best answers provided by users) to find expertsfor different categories. Zhou et al. extend PageRankalgorithm to a topic-sensitive probabilistic modelwhich finds the experts by taking into account boththe link structure and the topical similarity ausers. Pal et al. propose a probabilistic model toidentify existed and potential experts through users’selection bias of questions. Zhao et al.Topic modeling is the basis of many tasks (e.g., ques-tion retrieval and question routing) in CQA services.Many topic models have been proposed or employedin CQA services. For PLSA based models, Wu et al.propose an incremental PLSA model for continuouslyupdating when the system absorbs newquestions, which considers short-term and long-terminterests of users during the updating procedure. Xuet al. propose a Dual-Role model based on PLSA,which explicitly models the asker role and answererrole of users. For LDA based models, Ji et al. propose a LDA-based model for question retrieval,assuming that a question-answer pair shares the sametopic distribution but with different vocabularies. Guoet al.
PROPOSED SYSTEM:
The main contributions of this work are summarizedas follows:_ We present a multi-objective learning-to-rank approachfor question routing in CQA services,which is referred to as MLQR. In MLQR, the answeringpossibility and answer quality of routedusers are optimized simultaneously. To our bestknowledge, this work is the first attempt to tacklethe question routing problem in a way of multiobjectiveoptimization._ We propose a tagword topic model (TTM) whichexplicitly models tag-word co-occurrences andaggregates them on the corpus level. TTM canderive good topical representations for questions,which can benefit not only question routing butalso other tasks (e.g., question retrieval) in CQAservices._ We develop features for a question-user pair,which capture a user’s interest, activeness andexpertise for a question at both platform leveland thread level. In particular, the platform-levelfeatures summarize the intrinsic information ofa user from his/her history posts in the CQAplatform, while the thread-level features learnvarious user performance by measuring his/hercompetition with others in each thread answeredby him/her._ We conduct extensive experiments to evaluate theperformance of our MLQR on real-world datasets.Experimental results show that our MLQRachieves significant improvements over state-ofthe-artmethods in terms of both answering possibilityand answer quality.\
CONCLUSION
In this paper, we tackle the question routing problemin CQA services in a way of multi-objective optimization.In particular, we first formulate questionrouting as a multi-objective ranking problem, and thenpresent a multi-objective learning-to-rank approach forquestion routing (MLQR). In MLQR, we propose atagword topic model (TTM) which uses corpus-levelaggregation of tag-word combinations to relieve thedata sparsity problem of questions. Based on TTM, wedevelop two sets of features which capture a user’sinterest, activeness and expertise for a question atboth platform level and thread level. The platformlevelfeatures summarize a user’s information fromhis/her history posts in the CQA platform, while thethread-level features learn a user’s performance bymeasuring his/her competition with others in his/heranswered threads. These features are further used inan extended learning-to-rank algorithm, which canoptimize the objectives of question routing simultaneously.Experimental results on real-world datasetsshow that our MLQR can outperform state-of-the-artmethods in terms of both answering possibility andanswer quality.
REFERENCES
[1] J. Liu, H. Shen, and L. Yu, “Question quality analysis andprediction in community question answering services withcoupled mutual reinforcement,” TSC, 2015.
[2] B. Li and I. King, “Routing questions to appropriate answerersin community question answering services,” in CIKM, 2010,pp. 1585–1588.
[3] B. Li, I. King, and M. R. Lyu, “Question routing in communityquestion answering: putting category in its place,” in CIKM,2011, pp. 2041–2044.
[4] T. C. Zhou, M. R. Lyu, and I. King, “A classification-based approachto question routing in community question answering,”in WWW, 2012, pp. 783–790.
[5] Z. Ji and B. Wang, “Learning to rank for question routingin community question answering,” in CIKM, 2013, pp. 2363–2368.
[6] G. Zhou, K. Liu, and J. Zhao, “Joint relevance and answerquality learning for question routing in community QA,” inCIKM, 2012, pp. 1492–1496.
[7] L. Yang, M. Qiu, S. Gottipati, F. Zhu, J. Jiang, H. Sun, andZ. Chen, “Cqarank: jointly model topics and expertise incommunity question answering,” in CIKM, 2013, pp. 99–108.[8] C. J. Burges, “From ranknet to lambdarank to lambdamart: Anoverview,” 2010.
[9] D. M. Blei, A. Y. Ng, and M. I. Jordan, “Latent dirichletallocation,” the Journal of machine Learning research, vol. 3, pp.993–1022, 2003.
[10] X. Yan, J. Guo, Y. Lan, and X. Cheng, “A biterm topic modelfor short texts,” in WWW, 2013, pp. 1445–1456.