数据挖掘-Predicting Web Searcher Satisfaction with Existing Community-based Answers.pptx

;;;;;;;;;Problem Define;Query clarity features query length click statistics for the query query entropy “WH”--question word Query question match features TFIDF, BM25,KL-divergence language model, overlap between the query and question, co-occurrence statistics between the query and the given question from the click data. Answer quality features answer quality features features related to asker satisfaction ;;;Logistic Regression;Logistic Regression;Direct Approach;Composite Approach;Datasets; Human Labeling Amazon Mechanical Turk (MTurk) was used to collect human judgments on how an answer satisfies a search query. ;Methods Compared;NDCG metric;Direct vs. Composite Comparison;Analysis and Discussion;Answer Ranking for Queries;;In our future work, we plan to further investigate the types of queries that are likely to be satisfied by CQA pages. Another brunch of potential work is to develop semi-supervised or unsupervised methods to predict searcher satisfaction, as large numbers of human labels are hard to obtain. Finally, we intend to study searcher satisfaction with other types of community generated Web pages that possess interesting structure, such as Facebook pages (subject to appropriate privacy policies).;