Segmentation of search topics in query logs
A system and method are disclosed for identifying the tasks that a user is attempting to perform when submitting a series of search queries. A task may be categorized hierarchically into a mission and a goal in order to classify a series of queries and determine the user's intent. The user's mission may be an extended task and the user's goal may be a specific task. A pair of queries may be analyzed to determine whether those queries have the same mission and/or the same goal. For a series of queries, boundaries may be identified between the queries to identify where a user has changed her goal or changed her mission. The identification of goals, missions and boundaries may be automated based on an analysis of features of the queries that are compared.
1. A method for identifying boundaries between a plurality of search queries from a session, the method comprising:
receiving the plurality of search queries from a search log database;
analyzing at least one feature of each at least one successive pair of search queries of the plurality of search queries, wherein the analyzing comprises comparing the at least one feature of each query in the at least one successive pair and indicating whether there is change between the at least one successive pair;
identifying a goal boundary between the at least one successive pair of queries when the analyzing indicates a change between the at least one successive pairs of queries, wherein the goal boundary establishes segments of queries characterized by similar goals; and
clustering related segments of queries that are characterized by a similar goal.
2. The method of claim 1 wherein the change comprises a difference in the at least one feature between the successive pair.
3. The method of claim 1 wherein the at least one feature comprises an edit distance comparison between the one of the successive pairs.
4. The method of claim 3 wherein the change comprises a high value for a normalized value of the edit distance.
5. The method of claim 1 wherein the at least one feature comprises at least one of a temporal feature, a character prefix overlap, a word prefix overlap, a common words feature, a ratio of common words feature, a log likelihood ratio score, a probability that a query is the last query, a vector similarity, or a number of unique domains.
6. The method of claim 1 wherein the goal boundary indicates that each query from the one of the successive pairs of queries have different goals.
7. The method of claim 1 wherein the clustering further comprises
analyzing features from different segments of queries to determine whether the segments are characterized by the similar goal; and
combining segments of queries that are characterized by the similar goal.
8. The method of claim 1 further comprising:
identifying a mission boundary between one of the successive pairs when the feature analysis indicates a change between the one of the successive pairs, wherein the mission boundary establishes a segment of successive queries with a similar mission; and
clustering related segments of queries that have the similar mission.
9. The method of claim 8 wherein a goal is a subset of a mission.
10. The method of claim 1 wherein the search queries from the session comprises a set of queries that are searched over a fixed time period.