Abstract: The efficiency of information retrieval from web depends largely on the search engine results page (SERP) that obtained by searchers, especially the highlighting text. At present, the SERP of commercial search engines usually uses query terms highlighting strategy. However, the query words can be ambiguous and even contain noise, which may be incompletely consistent with the search intention of users. In order to highlight the most important terms that describe the search information clearly, this paper proposes a new key term highlighting strategy based on the results of manual annotation. Then this paper generates highlighting terms based on four machine learning algorithms, including structured support vector machine, hidden Markov model, max-margin Markov networks and conditional random field algorithm. In addition, this paper also proposes a new method which called the joint sequence labeling (JSL) algorithm to combine these four structured learning algorithms. Moreover, this paper conducts search experiments by using JSL algorithm. Experimental results show that the JSL algorithm provides more accurate solutions compared with the baselines and its search accuracy achieves 9330%. And the results of search experiments show that the key term highlighting strategy achieves better performance and users’satisfactory than traditional query terms highlighting strategy.

Key words: search engine results page (SERP); intent of user; query terms highlighting; joint sequence labeling (JSL) algorithm