1. Understanding User’s Query Intent with Wikipedia G200949016 여 승 후

2. content Abstract Introduction Wikipedia Methodology Experiments Conclusion

3. Abstract Understanding the intent behind a user’s query can help search engine to automatically route three major challenges to the query intent classification problem –Domain coverage – Intent representation –Semantic interpretation the statistical machine learning approaches is difficult and often requires many human efforts Propose a general methodology to the problem of query intent classification with Wikipedia

4. Introduction Intent representation challenge -how to define a semantic representation that can precisely understand and distinguish the intent of the input query Domain coverage challenge –how to clarify the semantic boundary of the intent Semantic interpretation challenge –how to correctly understand the semantic meaning of the input query The motivation of this work is to meet these challenges in query intent classification. Wikipedia, to help understand a user’s query intent.

5. Wikipedia - introduce the structure of Wikipedia Wikipedia ? multilingual, web-based, free content encyclopedia written collaboratively by more than 75,000 regular editing contributors 1.Article Links - Wikipedia is structured as an interconnected network of articles. contributor can insert a hyperlink between a word or phrase that occurs in the article and corresponding 2. Category Links - both articles and categories can belong to more than one category ex> “Puma” belongs to two categories: “Cat stubs” and “Felines”. 3.Redirect Links - redirect page exists for each alternative name

6. Wikipedia - introduce the structure of Wikipedia

7. Methodology - present a new methodology for understanding a user’s query intent using Wikipedia knowledge 1. select a few queries (should be representative queries of the specific domain ) 2.map these queries to Wikipedia concepts 3.browse the categories to which they belong (their sibling concepts and the concepts they link to in Wikipedia ) Result - easily collect a large amount of representative Wikipedia concepts

8. Experiments - Define and Propose three kinds of query intent applications and then show the results of several experiments 1.Travel Intent Identification 2. Personal Name Intent Identification 3. Job Intent Identification

9. Experiments Algorithms 1. Supervised learning with seed concepts. using logistic regression based on seed concepts selected from Wikipedia denote this method as “LR”. 2. Supervised learning with concepts expanded with Wikipedia. using logistic regression on selected concepts denote this method as “LRE” 3. Our method the intent predictor part and the random walk runs for 100 iterations denote this method as “WIKI” 4. Our method without the random walk step. denote this method as “WIKI-R”

10. Random walk algorithm –propagate intent from the seed examples into the Wikipedia ontology – assign an intent score to each Wikipedia concept – obtain an intent probability for each concept in Wikipedia

11. our method reaches the best performance in terms of F1 threshold of probability based on the tuning set Compared and Results

12. Conclusion the world’s largest knowledge resource, Wikipedia, to help understand a user’s query intent to minimize the human effort required to investigate the features of a specified domain and understand the users’ intent behind their input queries achieve by mining the structure of Wikipedia and propagating a small number of intent seeds through the Wikipedia structure achieves much better classification accuracy than other approaches.

13. Q & A