A Method For Query Intent Identification Based On Markov Network Clique

With the development of the Internet, data on the network increasesexponentially, to provide user more accurate search results with the informationneeded to become one of the hot issues in the field of information retrieval research.Traditional information retrieval system is simply matching the words in thedocument with the query words by a simple calculation, to get the list of relateddocuments with query, and this are often unable to provide satisfactory results to theuser. Because some of the words is ambiguous, and the meaning of the word may bechanging over time.How to organize the relationship between the query and document appropriately.we must first solve the problem between the words, because the query is often veryshort, so we need to use additional information to expand the query, the Markovnetwork is a good query expansion method, it is different from our traditional searchmethods, it is the technology of graph theory, computer science, probability theory,the ideological integration. Markov network is widely used for uncertain knowledgerepresentation and reasoning, as well as between variables transfer process, it is aneffective way to address the issue of uncertainty. Information retrieval system hasmany problems, because it can not accurately understand the user’s query intent. So,imagine if we can use Markov network to identify the query intent, and using theintent of the query for subsequent retrieval process, will be better to improveretrieval efficiency.Based on the idea above, we propose the method on query intent classification.We make use of manually labeled queries form Sogou’s query log (about2250) astraining data, and use the ten classes data to construct the Markov network. So wecan effectively get information of the queries. After this process we re-search thequeries in this data sets. Returning the relevance results of the queries, and classifythis results according the classifier trained by the ten classes data. At last, we canpredict the category of queries and retrieval the queries again. In the experiments weuse the11-avg and3-avg as our assessment process. Experiment results demonstratethe algorithm in this paper presents some advantages compared with other methods.