This project addresses the desktop search problem by considering various techniques for ranking results of a search query over the file system. We consider:

• basic ranking techniques, which are based on a single file feature (e.g., file name, file content, access date, etc.)
• learning-based ranking schemes
• a novel ranking technique, based on query selectiveness

The learning techniques have been shown to be significantly more effective than the basic ranking methods. Our ranking technique based on query selectiveness is effective for use during the cold-start period of the system, even though it does not involve any learning.

All our results are derived after a thorough user study and post-mortem analysis of the user log files.

People:

• Naama Zwerdling (Aharoni)
• Sara Cohen
• Carmel Domshlak

Papers

• On Ranking Techniques for Desktop Search (WWW07 poster paper)
• On Ranking Techniques for Desktop Search (Msc. Thesis)

Raw Data:

The raw data used for analysis is available here. The following fields are available:

• ResultID: Unique id given to each result file returned to each query.
• UserName: Unique id given to each user.
• QueryID: Unique id given to each query. Note that there may be many results with the same QueryId, since a query may return many results.
• Hit: True or false, depending on whether the user chose this particular result.
• ContentScore: Measures the similarity of the file content to the query.
• NameScore: Measures the similarity of the file name to the query.
• PathScore: Measures the similarity of the file path to the query.
• QueryLog: Measures the similarity of the previous queries for which this file was chosen to the current query.
• AccessedDiff: Difference in time between the last access date of the file and the time the query was issued.
• UpdatedDiff: Difference in time between the last update date of the file and the time the query was issued.
• CreatedDiff: Difference in time between the create date of the file and the time the query was issued.
• Size: Size of the file.
• Level: Distance of the file from the outermost directory.
• File Type: Type of file (i.e., file extension).
• DirRank: Directory-based ranking of the file.
• NormSize: Normalized size of the file. (Normalization is with respect to other files of the same type.)

For more details about these fields, see our papers.