Method and apparatus for calculating similarity among documents
View Patent ↗Information that individual elements (characteristic character strings) indicative of characteristics of a registered document appear in the registered document is stored in advance. When calculating similarity of the registered document, a query designated by a searcher is analyzed. The query is represented by a characteristic vector having the individual elements which take the relation between a plurality of words into consideration. Pieces of appearance information of the individual words contained in the query are counted. The counted appearance information is compared with a searching index to calculate similarity between documents.
1. A similarity calculation method for calculating similarity among documents in a document search system adapted to search documents registered in advance, comprising:
storing, in a storage as a searching index, each of keywords contained in a previously registered document;
extracting each of keywords contained in a query inputted by a searcher;
deciding an attribute of each of said keywords by referring to an element-type dictionary which provides correspondence relationships between keywords and attributes;
classifying said extracted keywords on an attribute basis in accordance with a result of decision and counting a number of keywords belonging to each of the attributes thus classified to determine a count number of each of the attributes;
comparing said count numbers of attributes thus determined with count numbers of corresponding attributes of said previously registered document;
calculating similarity between said query and said previously registered document on a basis of comparison of said count numbers; and
wherein the searching index stores character strings with respect to a plurality of previously registered documents;
wherein at least one keyword of the keywords contained in the query is decided as a first group of the keywords, and at least one different keyword of the keywords is decided as a second group of the keywords;
wherein the deciding, classifying, counting, comparing and calculating operations are performed for the first group of the keywords to determine a first subset of the previously registered documents having similarity to the at least one keyword, and are separately performed for the second group of the keywords to determine a second subset of the previously registered documents having similarity to the at least one different keyword;
the similarity calculation method further comprising:
determining any previously registered documents which are common to both the first and second subsets to narrow-down to documents having similarity to said query.
2. A computer-readable medium having computer-readable code therein, which, when extended by a computer, causes the computer to implement a similarity calculation method for calculating similarity among documents in a document search system adapted to search documents registered in advance, the similarity calculation method comprising:
storing, in a storage as a searching index, each of keywords contained in a previously registered document;
extracting each of keywords contained in a query inputted by a searcher;
deciding an attribute of each of said keywords by referring to an element-type dictionary which provides correspondence relationships between keywords and attributes;
classifying said extracted keywords on an attribute basis in accordance with a result of decision and counting a number of keywords belonging to each of the attributes thus classified to determine a count number of each of the attributes;
comparing said count numbers of attributes thus determined with count numbers of corresponding attributes of said previously registered document;
calculating similarity between said query and said previously registered document on a basis of comparison of said count numbers; and
wherein the searching index stores character strings with respect to a plurality of previously registered documents;
wherein at least one keyword of the keywords contained in the query is decided as a first group of the keywords, and at least one different keyword of the keywords is decided as a second group of the keywords;
wherein the deciding, classifying, counting, comparing and calculating operations are performed for the first group of the keywords to determine a first subset of the previously registered documents having similarity to the at least one keyword, and are separately performed for the second group of the keywords to determine a second subset of the previously registered documents having similarity to the at least one different keyword;
the similarity calculation method further comprising:
determining any previously registered documents which are common to both the first and second subsets to narrow-down to documents having similarity to said query.
3. A document search system adapted to search documents registered in advance, comprising:
a processor adapted with software and supportive hardware, for calculating similarity among documents, the processor adapted to effect operations comprising:
storing, in a storage as a searching index, each of keywords contained in a previously registered document;
extracting each of keywords contained in a query inputted by a searcher;
deciding an attribute of each of said keywords by referring to an element-type dictionary which provides correspondence relationships between keywords and attributes;
classifying said extracted keywords on an attribute basis in accordance with a result of decision and counting a number of keywords belonging to each of the attributes thus classified to determine a count number of each of the attributes;
comparing said count numbers of attributes thus determined with count numbers of corresponding attributes of said previously registered document;
calculating similarity between said query and said previously registered document on a basis of comparison of said count numbers; and
wherein the searching index stores character strings with respect to a plurality of previously registered documents;
wherein at least one keyword of the keywords contained in the query is decided as a first group of the keywords, and at least one different keyword of the keywords is decided as a second group of the keywords;
wherein the deciding, classifying, counting, comparing and calculating operations are performed for the first group of the keywords to determine a first subset of the previously registered documents having similarity to the at least one keyword, and are separately performed for the second group of the keywords to determine a second subset of the previously registered documents having similarity to the at least one different keyword;
the processor adapted to effect further operations comprising:
determining any previously registered documents which are common to both the first and second subsets to narrow-down to documents having similarity to said query.
4. A similarity calculation method for calculating similarity among documents in a document search system adapted to search documents registered in advance, comprising:
storing, in a storage as a searching index, each of keywords contained in a previously registered document;
extracting each of keywords contained in a query inputted by a searcher;
deciding an attribute of each of said keywords by referring to an element-type dictionary which provides correspondence relationships between keywords and attributes;
classifying said extracted keywords on an attribute basis in accordance with a result of decision and counting a number of keywords belonging to each of the attributes thus classified to determine a count number of each of the attributes;
comparing said count numbers of attributes thus determined with count numbers of corresponding attributes of said previously registered document; and
calculating similarity between said query and said previously registered document on a basis of comparison of said count numbers.
5. A computer-readable medium having computer-readable code therein, which, when executed by a computer, causes the computer to implement a similarity calculation method for calculating similarity among documents in a document search system adapted to search documents registered in advance, the similarity calculation method comprising:
storing, in a storage as a searching index, each of keywords contained in a previously registered document;
extracting each of keywords contained in a query inputted by a searcher;
deciding an attribute of each of said keywords by referring to an element-type dictionary which provides correspondence relationships between keywords and attributes;
classifying said extracted keywords on an attribute basis in accordance with a result of decision and counting a number of keywords belonging to each of the attributes thus classified to determine a count number of each of the attributes;
comparing said count numbers of attributes thus determined with count numbers of corresponding attributes of said previously registered document; and
calculating similarity between said query and said previously registered document on a basis of comparison of said count numbers.
6. A document search system adapted to search documents registered in advance, comprising:
a processor adapted with software and supportive hardware, for calculating similarity among documents, the processor adapted to effect operations comprising:
storing, in a storage as a searching index, each of keywords contained in a previously registered document;
extracting each of keywords contained in a query inputted by a searcher;
deciding an attribute of each of said keywords by referring to an element-type dictionary which provides correspondence relationships between keywords and attributes;
classifying said extracted keywords on an attribute basis in accordance with a result of decision and counting a number of keywords belonging to each of the attributes thus classified to determine a count number of each of the attributes;
comparing said count numbers of attributes thus determined with count numbers of corresponding attributes of said previously registered document; and
calculating similarity between said query and said previously registered document on a basis of comparison of said count numbers.