|
Many queries are short and query terms are noisy. As an example, in the KDDCUP 2005 dataset, queries containing 3 words are most frequent (22%). Furthermore, 79% queries have no more than 4 words. A user query often has multiple meanings. For example, "''apple''" can mean a kind of fruit or a computer company. "''Java''" can mean a programming language or an island in Indonesia. In the KDDCUP 2005 dataset, most of the queries contain more than one meaning. Therefore, only using the keywords of the query to set up a [[vector space model]] for classification is not appropriate.
* Query-enrichment based methods<ref>Shen et al. [http://www.sigkdd.org/sites/default/files/issues/7-2-2005-12/KDDCUP2005Report_Shen.pdf "Q2C@UST: Our Winning Solution to Query Classification"]. ''ACM SIGKDD Exploration, December 2005, Volume 7, Issue 2''.</ref><ref>Shen et al. [http://portal.acm.org/ft_gateway.cfm?id=1165776 "Query Enrichment for Web-query Classification"]. ''ACM TOIS, Vol. 24, No. 3, July 2006''.</ref> start by enriching user queries to a collection of text documents through [[search engines]]. Thus, each query is represented by a pseudo-document which consists of the snippets of top ranked result pages retrieved by search engine. Subsequently, the text documents are classified into the target categories using synonym based classifier or statistical classifiers, such as [[Naive Bayes]] (NB) and [[Support Vector Machines]] (SVMs).
=== How to adapt the changes of the queries and categories over time? ===
The meanings of queries may also evolve over time. Therefore, the old labeled training queries may be out-of-data and useless soon. How to make the classifier adaptive over time becomes a big issue. For example, the word "''Barcelona''" has a new meaning of the new micro-processor of AMD, while it refers to a city or football club before 2007. The distribution of the meanings of this term is therefore a function of time on the Web.
* Intermediate taxonomy based method<ref>Shen et al. [http://portal.acm.org/ft_gateway.cfm?id=1148196 "Building bridges for web query classification"]. ''ACM SIGIR, 2006''.</ref> first builds a bridging classifier on an intermediate taxonomy, such as [[Open Directory Project]] (ODP), in an offline mode. This classifier is then used in an online mode to map user queries to the target categories via the intermediate taxonomy. The advantage of this approach is that the bridging classifier needs to be trained only once and is adaptive for each new set of target categories and incoming queries.
=== How to use the unlabeled query logs to help with query classification? ===
Since the manually labeled training data for query classification is expensive, how to use a very large web search engine query log as a source of unlabeled data to aid in automatic query classification becomes a hot issue. These logs record the Web users' behavior when they search for information via a search engine. Over the years, query logs have become a rich resource which contains Web users' knowledge about the World Wide Web.
* Query clustering method<ref>Wen et al. [http://portal.acm.org/ft_gateway.cfm?id=503108 "Query Clustering Using User Logs"], ''ACM TOIS, Volume 20, Issue 1, January 2002''.</ref> tries to associate related queries by clustering "session data", which contain multiple queries and click-through information from a single user interaction. They take into account terms from result documents that a set of queries has in common. The use of query keywords together with session data is shown to be the most effective method of performing query clustering.
* Selectional preference based method<ref>Beitzel et al. [http://portal.acm.org/ft_gateway.cfm?id=1229183 "Automatic Classification of Web Queries Using Very Large Unlabeled Query Logs"], ''ACM TOIS, Volume 25, Issue 2, April 2007''.</ref> tries to exploit some [[association rules]] between the query terms to help with the query classification. Given the training data, they exploit several classification approaches including exact-match using labeled data, N-Gram match using labeled data and classifiers based on perception. They emphasize on an approach adapted from computational linguistics named selectional preferences. If x and y form a pair (x; y) and y belongs to category c, then all other pairs (x; z) headed by x belong to c. They use unlabeled query log data to mine these rules and validate the effectiveness of their approaches on some labeled queries.
== Applications ==
| 