Research Summary for 2013-2014

My research interest generally lies in the intersection between data mining and user modeling. I am especially interested in user understanding, knowledge discovery and decision optimization, via designing novel computational models. My works have result in over 20 publications in top venues in data mining and information retrieval areas, including SIGIR, KDD, WWW, WSDM and CIKM.
In particular, my Ph.D. research in the year of 2013 and 2014 aims at developing general and effective computational models to: 1) discover knowledge via broad exploration of user search behaviors recorded in search engine logs; 2) optimize the automated knowledge service systems' output via the learned user models when interacting with the users. (Slides) (Poster)
  1. User Modeling in Search Logs via A Nonparametric Bayesian Approach [WSDM'14a]

    In this work, we study the problem of user modeling in the search log data and propose a generative model, dpRank, within a non-parametric Bayesian framework. By postulating generative assumptions about a user's search behaviors, dpRank identifies each individual user's latent search interests and his/her distinct result preferences in a joint manner. Experimental results on a large-scale news search log data set validate the effectiveness of the proposed approach, which not only provides in-depth understanding of a user's search intents but also benefits a variety of personalized applications.
    Fig 1. Latent user group: a homogenous unit of query and clicks Fig 2. User: a heterogeneous mixture over the latent user groups
    Fig 3. Generation of latent user groups: Dirichlet Process priors Fig 4. Another layer of DP to support infinite mixture of latent user groups
  2. Personalized Ranking Model Adaptation for Web Search [SIGIR'13]

    In this paper, we propose a general ranking model adaptation framework for personalized search. Using a given user-independent ranking model trained offline and limited number of adaptation queries from individual users, the framework quickly learns to apply a series of linear transformations, e.g., scaling and shifting, over the parameters of the given global ranking model such that the adapted model can better fit each individual user's search preferences. Extensive experimentation based on a large set of search logs from a major commercial Web search engine confirms the effectiveness of the proposed method compared to several state-of-the-art ranking model adaptation methods.
    Fig 5. Adjusting the generic ranking model's parameters with respect to each individual user's ranking preferences
  3. Adapting Deep RankNet for Personalized Search [WSDM'14b]

    In this paper, we first continue-trained a variety of RankNets with different number of hidden layers and network structures over a previously trained global RankNet model, and observed that a deep neural network with five hidden layers gives the best performance. To further improve the performance of adaptation, we propose a set of novel methods categorized into two groups. In the first group, three methods are proposed to properly assess the usefulness of each adaptation instance and only leverage the most informative instances to adapt a user-specific RankNet model. These assessments are based on KL-divergence, click entropy or a heuristic to ignore top clicks in adaptation queries. In the second group, two methods are proposed to regularize the training of the neural network in RankNet: one of these methods regularize the error back-propagation via a truncated gradient approach, while the other method limits the depth of the back propagation when adapting the neural network. We empirically evaluate our approaches using a large-scale real-world data set. Experimental results exhibit that our methods all give significant improvements over a strong baseline ranking system, and the truncated gradient approach gives the best performance, significantly better than all others.
    Fig 6. Adjusting the generic ranking model's parameters with respect to each individual user's ranking preferences

Publications

  1. Hongning Wang, Anlei Dong and Yi Chang. Joint Learning Approach from Clickthroughs, book chapter in Bo Long and Yi Chang (eds), Relevance Ranking for Vertical Search Engines, Morgan Kaufmann Publisher, 2014, p10-26.
  2. Hongning Wang, ChengXiang Zhai, Feng Liang, Anlei Dong and Yi Chang. User Modeling in Search Logs via A Nonparametric Bayesian Approach. The 7th ACM Web Search and Data Mining Conference (WSDM'2014). (PDF)
  3. Yang Song, Hongning Wang and Xiaodong He. Adapting Deep RankNet for Personalized Search. The 7th ACM Web Search and Data Mining Conference (WSDM'2014). (PDF)
  4. Hongning Wang, Xiaodong He, Ming-Wei Chang, Yang Song, Ryen White and Wei Chu. Personalized Ranking Model Adaptation for Web Search. The 36th Annual ACM SIGIR Conference (SIGIR'2013), p323-332, 2013. (PDF, slides)