COMPARATIVE WEB SEARCH SYSTEM AND METHOD

• US 20080222140A1
• Filed: 02/20/2008
• Published: 09/11/2008
• Est. Priority Date: 02/20/2007
• Status: Active Grant

• Current Assignee: Wright State University
• Original Assignee: Wright State University
• Inventors: Dong, Guozhu, Lagad, Hardik H.
• Granted Patent: US 7,912,847 B2
• Time in Patent Office: Days
• US Class Current: 707/5
• CPC Class Codes: G06F 16/951 Indexing; Web crawling techniques

Abstract

A system and method for a comparative web search engines, search result summarization, web snippet processing, comparison analysis, information visualization, meta-clustering, and quantitative evaluation of web snippet quality are disclosed. The present invention extends the capabilities of web searching and informational retrieval by providing a succinct comparative summary of search results at either the object or thematic levels.

Claims

1. A method performing a comparative web search comprising:

providing a meta-search engine in communication with a plurality of web-based search engines;

providing said meta-search engine with a query, a search mode, and a selected set of said web-based search engines, said meta-search engine using said query to search for documents on the selected set of said web-based search engines;

retrieving automatically search results from each of the web-based search engines in the selected set in the form of at least web snippets or documents from each member of the selected set of said web-based search engines and using the search result as raw data;

providing automatically the raw data to a data pre-processing module which automatically removes stop words and HTML tags, and applies a stemming algorithm, resulting in pre-processed data;

providing automatically the pre-processed data to a comparison engine, said comparison engine performing an object level comparison or a thematic level comparison depending on which comparison is specified in the search mode, said comparison resulting in a plurality of result sets from the selected set of said web-based search engines;

determining automatically logical relationships between each of the plurality of result sets and providing a results comparison of the determined logical relationships;

organizing automatically the search results in ranked lists when the object level comparison is performed by the comparison engine and labeled hierarchical clusters when the thematic level comparison is performed by the comparison engine, said organizing resulting in organized search results; and

outputting the results comparison and the organized search results for viewing.

2. The method of claim 1, wherein the results comparison is outputted on a graphical user interface as a Venn diagram.

3. The method of claim 1, wherein when the thematic level comparison is performed, the comparison engine uses selection of keywords from the web snippets to determine theme(s) of the result sets for each member of the selected set to establish common themes between the two or more result sets.

4. The method of claim 1, wherein when the object level comparison is performed, the comparison engine uses URL exactness to determine common results between the result sets for each member of the selected set which comprises removing a protocol part of the URL and eliminating unnecessary characters in the URL.

5. The method of claim 1, further comprises determining automatically appropriate keyword(s) for each of the labeled hierarchical clusters when the thematic level comparison is performed by the comparison engine.

6. The method of claim 1, further comprises determining automatically appropriate keyword(s) for each of the labeled hierarchical clusters when the thematic level comparison is performed by the comparison engine, which comprises comparing descriptions of the web snippets within each of the clusters to derive a meaningful cluster label for each of the clusters.

7. The method of claim 1, wherein when the thematic level comparison is performed, the comparison engine uses selection of keywords from the web snippets to determine theme(s) of the result set for each member of the selected set to establish common themes between the two or more result sets, which comprises selection of keyword(s) from each result in every one of the result sets, forming base clusters from corresponding ones of the result sets such that each base cluster comprises exactly one unique result, performing homogeneous clustering to identify overlap in keywords between two base clusters belonging to the same result set, performing heterogeneous clustering to identify overlap in keywords between two clusters belonging to different result sets, and performing cluster labeling on results of the heterogeneous and homogeneous clustering.

8. The method of claim 1, further comprises determining automatically appropriate keyword(s) for each of the labeled hierarchical clusters when the thematic level comparison is performed by the comparison engine, said determining comprises removing from each of the clusters stop words, words with document frequencies above a predetermined frequency, and terms of the query to provide an initial word set for each associated cluster, scoring the initial word set based on term frequency and appearance of the word in title, URL and/or in description of web result snippet to determine the final word set, and using the words from the final word set as the appropriate keyword(s) for the associated cluster.

9. The method of claim 1, wherein when the thematic level comparison is specified in the search mode, said method further comprises generating automatically a set of base clusters from the raw data collected from each member of the selected set of the generic search engine, performing homogeneous clustering to identify overlap in keywords between two base clusters belonging to the same result set, and performing heterogeneous clustering to identify overlap in keywords between two clusters belonging to different result sets.

13. The method of claim 1, wherein the logical relationships determined by the comparison engine is an intersection of results common to all of the plurality of result sets, results common to at least one subset of the plurality of result sets, and results unique to each of the plurality of result sets.

14. The method of claim 1, wherein the logical relationships determined by the comparison engine is an intersection of results common to all of the plurality of result sets, and results unique to each of the plurality of result sets.

15. The method of claim 1, further comprises performing a comparison by the comparison engine between data sets collected in response to a common query performed at different points of time.

16. The method of claim 1, further comprises performing a comparison by the comparison engine between the plurality of result sets which represents a current result set and a previously saved result set in response to a common query performed at a different point of time, said comparison yielding a comparative summary showing at least one of vanished results which are results uniquely found in the previously saved result set, new results which are results uniquely found in current result set, steady results which are results found in both the current and previous result sets at approximately same ranks, emerging results which are results found in both result sets where rank of a result in current result set is higher than its rank in previously saved result set, and disappearing results which are results found in both result sets where the rank in previously saved result set is higher than its rank in current set.

17. The method of claim 1, further comprises performing quantitative evaluation of web snippet quality providing at least a listing of non-query term(s) with highest document frequency that are very closely related to the query, and further providing a user selected parameter which when selected outputs an integer value indicating number of clusters with distinct themes formed from results (N+1 . . . 2N) retrieved from the search engine used in the selected set in response to the query.

18. The method of claim 1, wherein the search mode selectable includes the following modes:

(a) a comparison of data sets collected from same or different generic web search engines in response to two or more different queries, (b) a comparison of data sets collected from same or different generic web search engines in response to a common query performed at different points of time, (c) a comparison of data sets collected from different generic web search engines in response to a common query performed simultaneously, (d) a comparison of data sets collected from same or different generic web search engines in response to a common query performed in different languages, (e) a comparison of result sets collected from same or different generic web search engines in response to a common query where intra-domain similarity between results of the same set is 100% while inter-geographic origin similarity between result sets is zero, (f) a comparison of result sets collected from same or different generic web search engines in response to a common query where intra-geographic origin similarity between results of same set is 100% while inter-geographic origin similarity between result sets is zero, and (g) a result set retrieved from a generic web search engine in response to a query is segmented into bins of equi-distant and equi-weighted segments and the segments are compared to generate comparative summaries.

19. The method of claim 1, wherein the comparison engine conducts one of the following four procedures based on the search mode:

i. determining common results between two result sets, clustering the common results based on themes they share, and ranking of results in each participating set to determine at least emerging, disappearing and steady results;

ii. combining the results from all the participating result sets into a data reservoir and then clustering them based on common themes;

iii. clustering the results belonging to the same set based on common themes, and then comparing the themes presented by clustered results in one result set with other sets to find commonalities;

or

iv. using term frequencies of keywords common to all participating result sets to determine at least what themes are emerging, disappearing or remaining steady.

20. A system implementing the method of claim 1.

10. The method of claim 10, wherein the cluster labeling comprises finding at least two base clusters with maximum overlap between keywords and merging keywords if one of the at least two base clusters is a subset of ones of the at least two base clusters.

11. The method of claim 10, wherein the cluster labeling comprises finding at least two base clusters with maximum overlap between keywords and merging keywords if one of the at least two base clusters is a subset of ones of the at least two base clusters, and wherein if there is a tie between the at least two base clusters in regards to sharing same number of overlapping keywords, then employing a tie breaking technique to break the tie therebetween for determining which keyword terms to use for the cluster labeling for the al least two base cluster with maximum overlapping.

12. The method of claim 11 further comprises finding at least two clusters with maximum overlap between their associated keywords and merging the keywords if one of the at least two clusters is a subset of ones of the at least two clusters, and wherein if there is a tie between the at least two clusters in regards to sharing the same number of overlapping keywords, then employing a tie breaking technique to break the tie therebetween for determining which keyword terms to use for cluster labeling of the al least two base clusters with maximum overlapping, said tie breaking technique comprises comparing the frequency scores of each of the at least two base clusters, wherein the highest score breaks the tie, and if the frequency scores are unable to break the tie, then checking if the keywords of each of the at least two clusters is a submersion of the another keywords, and if still no resolution to the tie, skipping merger and labeling each of the at least two clusters with the associated keywords.