1. SIGIR’09 Boston 1 Entropy-biased Models for Query Representation on the Click Graph Hongbo Deng, Irwin King and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong Kong July 21st, 2009

2. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 2 Introduction Query suggestion Query classification Targeted advertising Ranking  Query log analysis – improve search engine’s capabilities

3. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 3 Introduction  Click graph – an important technique A bipartite graph between queries and URLs Edges connect a query with the URLs Capture some semantic relations, e.g., “map” and “travel” How to utilize and model the click graph to represent queries? Robustness: Some queries with skewed click count may exclusively influence the click graph Spam: Raw CF can be easily manipulated Traditional model based on the raw click frequency (CF) Propose an entropy-biased framework

4. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 4 Motivation  Basic idea Various query-URL pairs should be treated differently  Intuition Common clicks on less frequent but more specific URLs are of greater value than common clicks on frequent and general URLs Is a single click on different URLs equally important? General URL Specific URL

5. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 5 Outline  Introduction  Related Work  Methodology Preliminaries Click Frequency Model Entropy-biased Model  Experiments  Conclusion

6. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 6 Modeling queries and URLs Click entropy & result entropy Related Work  Using click graph Query clustering (Befferman and Berger, KDD’00, Wen et al., WWW’ 01) Random walks for relevance rank in image search (Craswell and Szummer, SIGIR’05) Query suggestion by computing the hitting time on a click graph (Mei et al., CIKM’08) Query classification from regularized click graph (Li et al., SIGIR’08) Using click graph These methods are proposed based on the click graph, while our objective is to investigate a better model to utilize and represent the click graph. Using click graph

7. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 7 Modeling queries and URLs Click entropy & result entropy Related Work  Using click graph Query clustering (Befferman and Berger, KDD’00, Wen et al., WWW’ 01) Random walks for relevance rank in image search (Craswell and Szummer, SIGIR’05) Query suggestion by computing the hitting time on a click graph (Mei et al., CIKM’08) Query classification from regularized click graph (Li et al., SIGIR’08) Using click graph These existing methods do not distinguish the variation on different query-URL pairs Using click graph  Modeling the representation Use the content of clicked Web pages to define a term-weight vector model for a query (Baeza-Yates et al., 2004) Represent query as a vector of documents (URLs) without considering the content information (Baeza-Yates and Tiberi, KDD’07) Propose the query-set document model to represent documents by mining frequent query patterns rather than the content information of the documents (Poblete et al., WWW’08) Modeling queries and URLs

8. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 8  Using click graph Query clustering (Befferman and Berger, KDD’00, Wen et al., WWW’ 01) Random walks for relevance rank in image search (Craswell and Szummer, SIGIR’05) Query suggestion by computing the hitting time on a click graph (Mei et al., CIKM’08) Query classification from regularized click graph (Li et al., SIGIR’08) Modeling queries and URLs Click entropy & result entropy Related Work Using click graph These methods are focused on personalization for different queries, while our entropy- biased models are focused on the weighting scheme of various query-URL pairs Using click graph  Modeling the representation Use the content of clicked Web pages to define a term-weight vector model for a query (Baeza-Yates et al., 2004) Represent query as a vector of documents (URLs) without considering the content information (Baeza-Yates and Tiberi, KDD’07) Propose the query-set document model to represent documents by mining frequency query patterns rather than the content information of the documents (Qin et al., WWW’08) Modeling queries and URLs  For personalization Explore click entropy to measure the variability in click results (Dou et al., WWW’ 07) Propose result entropy to capture how often results change (Teevan et al., SIGIR’08) Click entropy & result entropy

9. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 9 Outline  Introduction  Related Work  Methodology Preliminaries Click Frequency Model Entropy-biased Model  Experiments  Conclusion

10. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 10 Preliminaries Query: URL: User: Query instance:

11. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 11 Traditional Click Frequency Model  Edges of click graph: Weighted by the raw click frequency (CF)  Transition probability Normalize CF From query to URL:From URL to query: Based on the transition probabilities, the query and document can be represented by the vector of transition probabilities respectively.

12. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 12 Traditional Click Frequency Model  Measure the similarity between queries The most similar query  q2 (“map”)  q1 (“Yahoo”) More reasonable  q2 (“map”)  q3 (“travel”) Cosine similarity: The CF model only considers the raw click frequency, and treats different query-URL pairs equally, even if some URLs are heavily clicked.

13. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 13 Methodology M Traditional click frequency model Entropy-biased models

14. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 14 Entropy-biased Model  The more general and highly ranked URL Connect with more queries Increase the ambiguity and uncertainty  The entropy of a URL: Suppose Tend to be proportional to the n(d j ) It would be more reasonable to weight these two edges differently because of the variation of the connected URLs.

15. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 15 Entropy  Discriminative Ability  Entropy increase, discriminative ability decrease Be inversely proportional to each other A URL with a high query frequency is less discriminative overall  Inverse query frequency Measure the discriminative ability of the URL Benefits  Constrain the influence of some heavily-clicked URLs  Balance the inherent bias of clicks for those highly ranked  Incorporate with other factors to tune the model

16. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 16 CF-IQF Model  Incorporate the IQF with the click frequency  A high click frequency  A low query frequency  “A” is weighted higher than “B”

17. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 17 CF-IQF Model  Transition probability The most similar query q2 (“map”)  q3 (“travel”) The most similar query q2 (“map”)  q1 (“Yahoo”)

18. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 18 UF Model and UF-IQF Model  Drawback of CF model Prone to spam by some malicious clicks (if a single user clicks on a certain URL thousands of times)  UF model Weight by user frequency instead of click frequency Improve the resistance against malicious click  UF-IQF model

19. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 19 Connection with TF-IDF  TF-IDF has been extensively and successfully used in the vector space model for text retrieval  Several researchers have tried to interpret IDF based on binary independence retrieval (BIR), Possion, information entropy and LM  TF-IDF has never been explored to bipartite graphs, and the IQF is new. The CF-IQF is a simplified version of the entropy-biased model  The entropy-biased model is employed to identify the edge weighting of the click graph, which can be applied to other bipartite graphs

20. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 20 Mining Query Log on Click Graph Query clustering Query-to-query similarity Query suggestion Models Query-to-query similarity Query suggestion

21. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 21 Similarity Measurement  Cosine similarity  Jaccard coefficient  The similarity results are reported and analyzed

22. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 22 Graph-based Random Walk  Query-to-query graph The transition probability from q i to q j  The personalized PageRank

23. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 23 Outline  Introduction  Related Work  Methodology Preliminaries Click Frequency Model Entropy-biased Model  Experiments  Conclusion

24. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 24 Experimental Evaluation  Data collection AOL query log data  Cleaning the data Removing the queries that appear less than 2 times Combining the near-duplicated queries 883,913 queries and 967,174 URLs 4,900,387 edges

25. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 25 Distributions

26. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 26 Evaluation: ODP Similarity  A simple measure of similarity among queries using ODP categories (query  category) Definition: Example:  Q1: “United States”  “Regional > North America > United States”  Q2: “National Parks”  “Regional > North America > United States > Travel and Tourism > National Parks and Monuments”  Precision at rank n (P@n):  300 distinct queries 3/5

27. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 27 Experimental Results  Query similarity analysis 1. CF-IQF is better than CF UF-IQF > UF The results support our intuition of the entropy-biased framework about treating various query-URL pairs differently 2. UF is better than CF UF-IQF > CF-IQF The results indicates the user frequency associated with the query-URL pair is more robust than the click frequency for modeling the click graph. Results:

28. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 28 Experimental Results  Query similarity analysis 3. TF-IDF is better than TF The improvements of CF-IQF over CF and UF-IQF over UF models are consistent with the improvement of TF-IDF over TF model. The reason: they share the same key point to identify and tune the importance of a term or a query- URL edge.

29. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 29 Experimental Results  Query similarity analysis 4. Jaccard coefficient The improvements are consistent with the Cosine similarity

30. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 30 Experimental Results  Query similarity analysis 5. UF-IQF achieves best performance in most cases. It is very essential and promising to consider the entropy-biased models for the click graph. 6. CF and UF models > TF CF-IQF, UF-IQF > TF-IDF The click graph catches more semantic relations between queries than the query terms

31. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 31 Experimental Results  Random Walk Evaluation Results: 1. With the increase of n, both models improve their performance. 2. CF-IQF model always performs better than the CF mode.

32. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 32 Experimental Results  Random Walk Evaluation In general, the results generated by the CF and the CF-IQF models are similar, and mostly semantically relative to the original query, such as “American airline”. Another important observation is that the CF- IQF model can boost more relevant queries as suggestion and reduce some irrelevant queries.

33. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 33 Conclusions  Introduce the inverse query frequency (IQF) to measure the discriminative ability of a URL  Identify a new source, user frequency, for diminishing the manipulation of the malicious clicks  Propose the entropy-biased models to combine the IQF with the CF as well as UF for click graphs  Experimental results show that the improvements of our proposed models are consistent and promising

34. Hongbo Deng, Irwin King, and Michael R. Lyu Department of Computer Science and Engineering The Chinese University of Hong KongSIGIR’09 Boston 34 Q&A Thanks!