JPH1021250A - Method for retrieving plural data bases and method for searching document between plural data bases - Google Patents

Abstract

PROBLEM TO BE SOLVED: To retrieve plural collections through a single path by calculating a score concerning each document through the use of statistics and information and processing all the related scores in the same mode regardless of the difference in search engines by a client. SOLUTION: Each of search engines 20 to 40 accesses a related data base 19 and may be provided with a unique arithmetic characteristic such as Boolean logic, a statistical inference. Each data base 19 generates a report 21 including the number N of records within the data base. This report 21 includes the number of appearing times of each searching word into a document responding to a query. Next, a report 25 generates the document identification number of each document including a hit with the number of the appearing times of each searching word. The score concerning each document is calculated 28 through the use of its own algorithm by a client software from this information.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

TECHNICAL FIELD This invention relates to searching and searching documents, and more particularly to those over a network.

[0002]

BACKGROUND OF THE INVENTION For over two decades, information services have provided access in multiple databases. For example, currently Knight-Ridder Information, Inc.
Dialog Information, also known as
Services (Dialog Information Services) provide hundreds of databases (also known as collections) available to searchers. Some of these databases contain bibliographic summaries, while others contain full-text documents. Searchers can apply queries to one or more databases. Initially, the searcher selects individual databases of interest based on past experience, or selects a group of databases selected by an information provider and related to a particular topic. For example, a searcher might select the topic patent. The information service on this topic has compiled several databases limited to patents. When a query is applied to a group of databases, the information service searches for the number of hits in each database. The searcher then accesses the database of interest and browses the individual records.
The system was originally designed for librarians and professional researchers who knew where to look for the desired information.

[0003] With the availability of wide area networks such as the Internet, not only search professionals, but also amateur users, have new opportunities in searching. New types of information providers have emerged that use public as well as private databases to provide bibliographic research data and literature to users. If a user is interested in a topic, such as a patent, the user may not know what resources they can collect to perform a search, nor the location of those resources. Since resources are often changed, the searcher will have less interest in the source of the response compared to the appropriateness of the response. It is recognized that distributed collections available over a wide area network can be treated as a single collection. Each sub-collection is searched individually and the reports are combined into a single list. It is also known that literature can be ranked and weighted by search engines according to algorithms, taking into account the nature of the particular collection. The score of the literature is
The collection of individual documents can be normalized to obtain the score that would be obtained if merged into a single integrated collection.

[0004] One of the problems with the prior art is that the score for each document is not absolute, but depends on the statistics in each collection and the algorithms associated with the search engine. The second that exists
The problem with is that the standard prior art procedure requires two passes. In the first pass, statistics are gathered from each search engine to calculate the weight for each query term. In the second step, the information from the first step is sent back to each search engine,
The search engine then assigns a specific weight or score to each hit or identified document. A third problem that exists is that the prior art requires that all collections use the same search engine.

[0005] It is an object of the present invention to use consistent criteria so that if the same document appears in two different databases, it will be scored the same when the results are merged. The idea is to devise a method for searching multiple collections in a single pass, with literature ranking. It is not necessary to use the same search engine for all collections.

[0006]

SUMMARY OF THE INVENTION The above objects are accomplished in a document search and search method that requires each participating search engine server to return statistics on the terms of each query in each of the returned documents. ing. The final relevance score is then calculated at the client terminal, not at the server. In this manner, all relevant scores are processed at the client in the same manner, independent of search engine differences.

[0007]

BEST MODE FOR CARRYING OUT THE INVENTION Referring to FIG. 1, the query indicated by query block 11 is specified by a user and provided to a terminal or client system. This query is sent electronically to the network interface 13. A network interface is an information service that has access to a source 17 that has a database related to the subject of the query. These databases deployed on other servers are polled simultaneously via communication channel 15. Communication channel 15 may be a wide area network link to source 17. The Internet is one model for such wide area network links and remote sources. The query is provided to a search engine. These search engines are in column 2
Represented at 0, 30, and 40, each search engine accesses an associated database at block 19. Each search engine may have its own computational characteristics, such as Boolean logic, statistical inference, and the like. Each database generates a report, indicated by block 21, which includes the number N of records in the database. The report also includes the number of times each search term appears in the literature responding to the query. This quantity, _NTERM, is indicated by block 23. Third, the report is block 25
, A document identification number of each document including a hit is generated together with the number of times each search word appears. From this information, the score calculation for each document is performed by the client software using its own algorithm, as indicated by block 27. For example, the formula for calculating the score is:

[0008]

(Equation 1)

Here, t_f= The time the word appears in the literature
A number, idf = log (N / N _TERM) And this
Case N and N_TERMIs reported by all collections
N and N announced_TERMIt is the sum of the values. Ipad score
Is sent to the output buffer indicated by block 29.
This output buffer is reserved for the person querying.
Calculate the top M scores that were required
Shift from 7. Through this database, only one
Note that only one pass was made. Total
The calculated score is treated as an absolute value.

[0010] In an alternative embodiment, optional parameters may be reported for use in the algorithm. Block 26 indicates that the frequency of the most frequent words in each document is reported for the purpose of using the client's more sophisticated ranking formula. Another optional data reduction step is that each search engine may calculate a score for the relevance of the document, in a manner known in the prior art. From this data, the search engine could pre-select up to the top M hits in the database. Here, M is the maximum number of hits requested by the user.

As an example, a search query may involve documents with the term "graphical user interface". Table 1 below shows a report generated by a search engine that selected some of the highest ranked documents. This report is returned to the user's client software, where the user uses the word frequency data and literature frequency data returned by each search engine to apply an algorithm, such as in formula (1) above. Therefore, a local calculation of document weights is performed for each query term, and _NTERM and N (= number of documents) returned from each collection are combined. Therefore, the word weighting is as if these collections were a single collection,
It will be exactly the same. Scoring is completely consistent, even when different search engines participate in the search, and the same document that appears in two different collections
You will always receive the same score.

[0012]

[Table 1]

[Brief description of the drawings]

FIG. 1 is a block diagram of a system according to the present invention.

[Explanation of symbols]

 Reference Signs List 11 query 13 network interface 15 communication channel 17 source 19 database 20 search engine 30 search engine 40 search engine

Claims (9)

[Claims] 1. A method for searching a plurality of databases that are distributed and accessible to a client via one or more search servers, comprising: (a) from a client each database associated with each database; Providing a search query to the server; (b) obtaining statistics for each database from each server to the client; and (c) obtaining, from each server to the client, a document obtained as a result of applying the query to the database. Obtaining information; and (d) at the client, calculating a score for each document using the statistics and the information, wherein the calculated scores are as if these databases were combined as a single database. Applicable to all databases It appears as a thing,
How to search multiple databases. 2. The method of claim 1, wherein the statistics for the collection include a size of the collection in relation to a number of records. 3. The method of claim 1, wherein information about each document includes the number of times each search term has appeared in that document. 4. The method of claim 1, wherein information about each database includes a number of documents containing each search term. 5. A method for searching a plurality of databases that are distributed and accessible to a client via one or more servers, comprising: (a) accessing each database from a client; (B) providing a search query from a client to a server associated with each database; (c) obtaining statistics for each database at the client; and (d) obtaining a result of applying a query to the database at the client. Obtaining statistical information about the relevant document, and (e) calculating, at the client, a score for the relevant document using the statistics and the information, the score calculated for the certain document. Is
It is independent of the database in which it appears,
How to search multiple databases. 6. A method for searching for documents among a plurality of databases in response to a search query, comprising: (a) providing a search query to each database; and (b) determining the number of records for each database. Determining, for each of said databases, applying a search query and recording a number of hits for each search query term and an identification of a database record having at least one hit in said number of hits; Counting the records having at least one hit for each of said databases and for each query term; and (e) calculated using the results of steps (b), (c) and (d); Reporting to the user the associated score of each record with respect to the search query. ,
A method of searching for documents among multiple databases. 7. The method of claim 6, further comprising selecting several databases and having more than one search engine for the databases prior to providing the same search query for all databases. The described method. 8. By selecting some records to be reviewed among the several databases, wherein the some records have the highest relevant scores for the search query. 7. The method according to 7. 9. Prior to calculating the associated score,
9. The method of claim 8, further defined by pre-selecting some records.