JP2004118877A - Information filtering device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide information filtering for realizing retrieval using only knowledge in conformity with a generation by automatically deleting a retrieval condition and a retrieval word obsolescent in a profile. <P>SOLUTION: A text information retrieving part 14 adds the present time (the newest adapted time) to the retrieval condition and the retrieval word in the profile adapted to text information in a similarity calculation by calculating similarity of the profile and the text information. While, a user information control part 20 receives the profile of adding this newest adapted time from a user information analyzing part 12; compares the newest adapted time added to the retrieval condition and the retrieval word in the profile with the present time; deletes the retrieval condition and the retrieval word unadapted to any text information for a specific period from the profile; and stores the profile corrected in this way in a user information storage part 13. <P>COPYRIGHT: (C)2004,JPO

Description

The present invention relates to an information filtering device that selects a user's request / interest from an enormous number of text information and presents it to the user.

In recent years, with the spread of word processors and electronic computers, and the spread of e-mail and electronic news via computer networks such as the Internet, the digitization of documents is accelerating. As the term "electronic publishing" implies, it is expected that information on newspapers, magazines and books will be provided electronically in the future. As a result, the amount of text information available to individuals in real time is expected to be enormous.

Along with this, there is an increasing demand for an information filtering system that selects a user's request / interest from a vast amount of text information such as newspapers and magazines and provides the user at any time.

Information filtering systems that have been implemented so far allow the user to evaluate the suitability of the presented text and feed the results back to search conditions called profiles, which search for text that interests the user. Some implement a relevance feedback function of improving the suitability of a text for each user.

However, in the future, information filtering systems will be used by multiple people who are interested in the same field, for example, research groups with specific themes in research laboratories. The relevance feedback in the conventional information filtering system is only for the purpose of responding to individual users, and it has not been possible to comprehensively analyze such requests of a plurality of users and to provide feedback. Also, text relevance judgments performed by individual users may not be consistent and reliable in some cases. Therefore, feedback does not always increase text relevance, and a more reliable relevance feedback function is used. Realization is desired.

Also, with conventional information filtering systems, new search terms can be added to the profile by relevance feedback, but the search terms that are no longer used in the text and are out of date are automatically deleted. Could not. Therefore, it is desired to realize an information filtering system that can change words and follow topics.

As described above, the relevance feedback in the conventional information filtering system is only for the purpose of responding to individual users, and it is not possible to comprehensively analyze such requests of a plurality of users and provide feedback. There was a problem.

Also, with conventional information filtering systems, new search terms can be added to the profile by relevance feedback, but the search terms that are no longer used in the text and are out of date are automatically deleted. Was not possible.

The present invention has been made in view of such circumstances, and automatically deletes outdated search conditions and search terms in a profile, thereby realizing a search that realizes a search using only knowledge in keeping with the times. The purpose is to provide filtering.

In order to achieve the above-described object, the present invention provides an information filtering device that selects desired text information from a plurality of text information and presents the text information to a user. Means for selecting the text information that matches the search condition, means for presenting the text information selected by the means to the user, and search conditions or search terms to which the text information presented by the means matches And a means for modifying the search condition based on the time information added by the means.

According to the present invention, time information is added to a search condition or a search word suitable for a search, and a profile of a user is corrected using the time information. As a result, for example, by automatically deleting from the profile search conditions and search terms that have not been matched for a long period of time, it is possible to always perform a search using highly compatible knowledge.

According to the present invention, there is provided an information filtering apparatus for selecting desired text information from a plurality of text information and presenting the selected text information to a user. Means for selecting the text information, means for presenting the text information selected by this means to the user, means for collecting feedback information that is the user's evaluation result on the text information presented by this means, First correction means for correcting the search condition based on the feedback information collected by this means, and time information added to a search condition or search word newly added at the time of correction by the first correction means. And second correcting means for correcting the search condition based on the time information added by this means. And butterflies.

In the present invention, time information is added to the newly added search condition or search word at the time of relevance feedback, and the user's profile is corrected using this time information. As a result, for example, a search using the latest knowledge can always be performed by automatically deleting from the profile search conditions and search terms that have been registered for a long period of time.

According to the present invention, it is possible to provide information filtering that realizes a search using only knowledge in keeping with the age by automatically deleting old search conditions and search words in a profile.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

(1st Embodiment)
First, a usage form of the information filtering system according to the first embodiment will be described with reference to FIG.

As shown in FIG. 1, in this embodiment, five users 3 (users A, B, C, D, and E) who have a common interest share the information filtering system 1. Then, text information arrives at the information filtering system 1 from the text information source 2 at any time. For example, if users A to E are a group conducting semiconductor research and register in the information filtering system 1 "we are interested in text information on semiconductors", this is converted into a search condition called a profile. You. Thereafter, the information filtering system 1 automatically extracts only the text information relating to the semiconductor from the new arrival information and presents the text information to the users A to E.

In the present embodiment, the information filtering system 1 has one common profile 10 for one group of users A to E, and narrows down information using this common profile 10 as a search condition. The text information presented to the users A to E is the same.

(System configuration)
FIG. 2 shows a functional configuration of the information filtering system 1 of the present embodiment. In the figure, solid arrows indicate the flow of data.

As illustrated, the information filtering system 1 includes a text information analysis unit 16, a text information storage unit 17, a text information search unit 14, a text information output unit 15, a user information input unit 11, a user information analysis unit 12, and a user information storage unit. It comprises a unit 13. Among these components, the text information analysis unit 12, the text information search unit 14, and the user information analysis unit 16 surrounded by a chain line can be realized by software executed by a central processing unit of a computer. The unit 13 and the user information storage unit 17 can be realized by a main storage device or a hard disk device of a computer. Further, the text information output unit 15 includes a CRT display or the like for presenting text information to the user 3. The user information input unit 11 includes a keyboard for inputting a topic of interest to the user 3 and relevance feedback information. And a mouse.

FIG. 3 shows an example of the processing flow of the text information analysis unit 16.
The text information analysis unit 16 first takes in text information from the text information source 2 (step A1). Here, the text information source 2 is an organization that generates text information and provides it to an information filtering system, such as a newspaper company or a publisher, or a separate entity that handles text information, such as an e-mail system or a document search system. Systems and sites that make text information publicly available on computer networks.

The text information analysis unit 16 performs morphological analysis, syntax analysis, semantic analysis, format analysis, and the like on the input text information, and performs frequency information and position information on text components such as words, phrases, sentences, and paragraphs, and text data. (5) A subject and 5W1H information are extracted (step A2).

(4) Each text is expressed by the extracted information (step A3). Subsequently, the information extracted from the text information is converted into a format that can be searched by the text information search unit 14 (step A4), and these are stored in the text information storage unit 17 (step A5). This corresponds to an indexing process in a normal information search.

FIG. 4 shows an example of the text information expressed by the text information analysis unit 16.
This figure is obtained when a newspaper article from XX Newspaper Company, which reads, "XX and XX companies will establish a semiconductor joint venture in XX Prefecture XX City on 17th of this month". It is an expression example of text information.

FIG. 5 shows an example of the processing flow of the text information search unit 14.
First, the text information search unit 14 extracts a profile, which is a search condition expressing the user's interest, from the user information storage unit 13 (step B1). Assuming that a plurality of users 3 have registered in the information filtering system 1 that they are "interested in information on semiconductors", for example, search conditions as shown in FIG. become.

FIG. 6 lists search conditions for searching for text information related to “semiconductor”. “(Condition 1)” determines whether or not the word “semiconductor” appears in the text. “(Condition 2)” indicates that the words “memory” and “semiconductor” This is to check whether they co-occur in the text. These correspond to the search conditions of a conventional pool search using a keyword. “(Condition 3)” is for searching for a text in which the name of a specific semiconductor company appears in the headline of the text, and is a search condition using information on the appearance position of a word and knowledge of the company name of the semiconductor company. ing. “(Condition 4)” is a search condition based on word frequency information. “(Condition 5)” is a search condition using a related word in the field of “semiconductor” such as “DRAM” or “flash memory”.

(6) The text information search unit 14 performs a text search using the profile 10 as shown in FIG. 6 as a search condition and the text stored in the text information storage unit 17 as a search target (step B2). Here, the text search corresponds to, for example, selecting text information that satisfies a search condition as shown in FIG. 6. Specifically, the text search is divided into text that satisfies the search condition and text that does not satisfy the search condition. This means that texts are ranked according to the degree of satisfying. For example, in the case of the latter, some of the top texts ranked by the search are selected for presentation by the user and passed to the text information output unit 15 (step B3). As a specific method of the text search, for example, a technique disclosed in a document (“SMART information search system”, edited by Gerald Sandton, translated by Shinji Kyoshin, Planning Center) may be adopted.

FIG. 7 shows an example of a processing flow of the text information output unit 15.
The text information output unit 15 receives text information to be presented to the user 3 from the text information search unit 14 (Step C1), and presents it to the user 3 (Step C2).

FIG. 8 shows an example of a processing flow of the user information input unit 11.
The user information input unit 11 receives user information from the user 3 (Step D1) and passes the user information to the user information analysis unit 12 (Step D2). The user information has the following two types.

The first is information on what text information the user 3 wants, such as "I want an article on semiconductors", which the user 3 pre-designates for the information filtering system 1. Here, this type of information is referred to as initial setting information.

The second is the relevance feedback information in which the user 3 has determined the suitability of the text information presented by the system. This is for modifying the profile 10 so that the article presented to the user 3 is more suitable for the request of the user 3. Specifically, for example, information in the form shown in FIGS. 9 and 10 is used. Can be considered.

FIG. 9 is an example of information obtained by performing a three-level evaluation of “necessary / slightly necessary / unnecessary” for each piece of text information presented by the user 3.
In this example, for example, a useful word is extracted from words included in “text 1” and “text 2” determined to be “necessary” and added to the profile 10, thereby performing processing such as the following. May have more filtering results.

As a modification of FIG. 9, the suitability may be determined not for each text but for each component of text. For example, it is conceivable to extract a sentence or paragraph of the text presented to the user 3 and feed back information such as “this part was useful” to the system. Further, in FIG. 9, three-stage evaluation is performed. However, the evaluation may be extended to evaluate suitability by numerical values.

FIG. 10 is an example of relevance feedback information in a natural language given by the user 3 via a keyboard or the like. Although an article related to “semiconductor manufacturing equipment” was presented at the top of the filtering result, if the user 3 is not very interested in “semiconductor manufacturing equipment”, a request as shown in FIG. From the next time, it is conceivable that the score of a text including the word "semiconductor manufacturing apparatus" may be reduced. In the case of a search method in which importance is assigned to search words in the profile 10, a request such as "put emphasis on DRAM over flash memory" is issued as shown in FIG. Thus, it is possible to change the importance of the search word in the profile.

FIG. 11 shows an example of a processing flow of the user information analysis unit 12.
The user information analysis unit 12 determines whether or not information has already been stored in the user information storage unit 13 (steps E1 and E2), and performs two operations according to the determination results. If the user information storage unit 12 is empty (Y in step E2), an initial selection information analysis process is performed (step E3), and if not, a relevance feedback information analysis process is performed (step E4).

FIG. 12 shows an example of the flow of the initial selection information analysis processing.
In the initial selection information analysis process, the user information analysis unit 12 analyzes the initial selection information with reference to an analysis dictionary 101 and the like necessary for language analysis prepared in advance, and extracts words and expressions representing the selected topic. Specify (step F1). Next, in order to obtain knowledge on a search word related to the selected topic and its synonyms, a search word is determined with reference to topic knowledge prepared in advance (step F2). Then, the profile 10 as shown in FIG. 6 is described using the determined search word. The generation of such a common profile 10 may be performed automatically by the system, or may be performed semi-automatically by, for example, causing the user 3 to modify the generated profile 10. The generated common profile 10 is stored in the user information storage unit 13 (Step F3).

FIG. 13 shows an example of the topic knowledge described above.
FIG. 13A is an example of topic knowledge describing the relationship between search terms. For example, since it is described that “ROM” and “RAM” are included in the lower concept of “semiconductor memory”, when describing the profile 10 for the topic “semiconductor memory”, “ROM” or “RAM” Can be used as a search term. FIG. 13B is an example of topic knowledge regarding synonym information. By utilizing such knowledge and registering not only “ROM” but also a synonym of “read only memory” as a search word in the profile, it is possible to perform a search with few misses.

FIG. 14 shows an example of the flow of the relevance feedback information breaking process.
In the relevance feedback information breaking process, the user information analysis unit 12 analyzes the relevance feedback information of each user 3 with reference to an analysis dictionary 101 and the like necessary for language analysis prepared in advance (step G1). -Step G4). Next, feedback information to be reflected on the profile 10 common to the user 3 is selected from among these pieces of information (step G5). Then, the common profile is updated using a relevance feedback method disclosed in, for example, a document ("SMART Information Retrieval System", edited by Gerald Sulton, translated by Shinji Kenko, Planning Center) and stored in the user information storage. It is stored in the unit 13 (step G6). When a new word is added at the time of updating the profile, the topic knowledge 102 on the topic may be referred to in order to obtain information on a related word of the word.

The difference between the conventional relevance feedback processing and the relevance feedback information analysis processing in this embodiment is that the former only reflects feedback information obtained from a single user in a single profile, while the latter That is, information to be reflected on the common profile 10 is selected from feedback information obtained from the user, and feedback is performed.

FIG. 15 shows an example of relevance feedback information obtained from a plurality of users.
In this example, three users 3 who are interested in the same topic share the information filtering system 1, and the relevance determination result for the three pieces of text information commonly presented to them is “O” or It is indicated by "x". Since the users A, B, and C are interested in the same topic, it is generally considered that the determination result of each user 3 matches, such as the compatibility determination result for “text 1” or “text 3”. . However, due to a slight discrepancy in the interests of the individual users 3 and a difference in knowledge in the field, or a difference in mood or busyness at the time of discrimination, a discrepancy appears as in the determination result of “Text 2” in the figure. Can be considered. In this example, the user A and the user B determine “text 2” as “useful”, but the user C determines “unnecessary”. In such a case, for example, by performing a feedback process assuming that “text 2” is useful by majority decision, it is considered that highly reliable feedback can be performed. Further, as a modified example, if information is given to the system in advance that the user A among the users A, B, and C in FIG. Relevance feedback can also be provided.

FIG. 16 shows a modification of the relevance feedback information obtained from a plurality of users.
Such feedback information can be obtained, for example, by extracting words frequently appearing in the text determined by each user 3 to be "useful" or a part thereof. In this example, the words “64 mega DRAM” and “semiconductor joint venture” frequently appear in the text specified by the user A or a part thereof. The same applies to user B and user C. Terms such as "semiconductor joint venture", "memory patent" and "semiconductor manufacturing equipment" are included only in feedback information of one user 3, whereas "64M DRAM" is used by all users. Included in feedback information. In such a case, “64 mega DRAM” is the most important information to be reflected in the common profile 10 because it is a word included in a text or a portion thereof that is determined to be useful by all users. it is conceivable that. Therefore, when reflecting only “64 Mega DRAM” on the common profile or reflecting all the words shown in FIG. 16 on the common profile 10, the weight of “64 Mega DRAM” is set higher than other words. If this is the case, highly reliable relevance feedback based on the majority decision of the plurality of users 3 can be performed. As a specific reflection method to the profile 10, for example, “64 mega DRAM” is added to “(condition 5)” in FIG. 6, or “(condition 1)” to “(condition 5)” is changed. For example, "64 Mega DRAM" may be added to all of them.

In the above, the relevance feedback based on the majority decision of the plurality of users 3 has been described. However, it is also possible to determine information to be reflected on the common profile 10 from the feedback information of the plurality of users 3 by other policies. For example, for each piece of text information, if there is at least one user 3 in the group who finds it useful, the text information may be used for feedback. In this case, in the example of FIG. 15, it can be determined that “text 1” and “text 2” are useful, as in the case of majority decision. It is considered that such a policy may be effective in a case where the group requires filtering that is as leak-free as possible as a whole. Similarly, for each text, if there is at least one user 3 who does not need it, it is conceivable to adopt a policy that the text is not used for relevance feedback. In this case, in the example of FIG. 15, only “text 1” is adopted as information suitable for the relevance feedback. Therefore, the user 3 may be allowed to sequentially designate the above-described policies, and the method of selecting information to be used for relevance feedback may be switched according to the designated policies.

(2nd Embodiment)
Next, a usage form of the information filtering system according to the second embodiment will be described with reference to FIG.
The difference between the first embodiment and the present embodiment is that the former has one profile 10 common to a plurality of users 3 and presents one filtering result using this to all the users 3. The latter has both a common profile 10 and a profile 18 for each user 3, and ultimately presents information customized for each user 3 to each user. The common profile 10 in the present embodiment reflects an information request common to all users 3, and the user profile 18 reflects an information request unique to the user 3. For example, even if users A to E in the figure are groups that are conducting research on the same theme, and first register in the information filtering system 1 as "I am interested in text information on semiconductors", a new text will be added over time. As information is taken in, there is a possibility that the demands of individual users 3 may change in small points. In the present embodiment, in order to deal with such a situation, the relevance feedback information obtained from the plurality of users 3 is divided into information to be reflected in the common profile 10 and information to be reflected in the user profile 18.

機器 The device configuration of this embodiment is the same as that of the first embodiment shown in FIG. The functions of the text information analysis unit 16 and the user information input unit 11 are the same as those described in the first embodiment. Here, only differences from the first embodiment will be described.

FIG. 18 shows an example of the processing flow of the text information search unit 14 in the present embodiment.
The difference between the present embodiment and the flow of the process of the text information search unit 14 of the first embodiment is that the latter performs a search using a combination of the common profile 10 and the individual user profiles 18 as a search condition, and the individual That is, a search result is obtained for each user 3. That is, for the user A, a search condition is created from the common profile 10 and the user profile 18 for the user A (step H4), and a search is performed according to the search condition (step H5), and the searched text is searched. The information is passed to the text information output unit 15 (Step H6). Similarly, for the user B, a search condition is created from the common profile 10 and the user profile 18 for the user B, a search is performed, and the searched text information is passed to the text information output unit 15. This is performed for all users 3.

FIG. 19 shows an example of the processing flow of the text information output unit 15 in the present embodiment.
The difference between the flow of processing of the text information output unit 15 of the first embodiment and this embodiment is that the latter outputs a filtering result for each user 3.

の う ち Of the processing flow of the user information analysis unit 12 in this embodiment, only the relevance feedback information analysis processing is different from that of the first embodiment, and this will be described below.

FIG. 20 shows an example of the flow of the relevance feedback information analysis processing in the present embodiment.
The difference between the flow of the relevance feedback information analysis processing of the first embodiment and the present embodiment is that in the latter, the information reflecting the relevance feedback information obtained from a plurality of users 3 in the common profile 10 and the individual user The point is that feedback is performed after sorting to information reflected in the profile 18. This will be described with an example in which feedback information as shown in FIG. 16 is obtained.

FIG. 16 shows three users 3 who are interested in “semiconductor” individually conducting feedback feedback and extracting words such as “64 mega DRAM” and “semiconductor joint venture” extracted from the information. I have. In this example, since the word "64 mega DRAM" appears in common among the feedback information of the three users, this word is a topic of "semiconductor" rather than a word expressing the individual users' detailed preferences. It may be useful information to modify the original search conditions for the search to be more current. For example, in the conventional text, only the word "16 mega DRAM" appeared, but when a new "64 mega DRAM" was developed and this word became commonly used in the text, is there. In such a case, the word "64 mega DRAM" is newly registered in the common profile 10, that is, the original search condition for obtaining general text information on "semiconductor". More generally, words commonly appearing in feedback information of many users 3 are used for updating the common profile 10. For example, of the words shown in FIG. 16, words appearing in the feedback information of two or more people are used for feedback to the common profile 10, and the remaining words are used for feedback to the individual user profiles 18. Only the "64 mega DRAM" is fed back to the common profile 10 (step J5).

(4) After the feedback on the common profile 10 is performed, a feedback process on each user profile 18 is performed (step J7). In the example illustrated in FIG. 16, the term “semiconductor joint venture” is obtained as feedback information unique to the user A. Therefore, the “semiconductor joint venture” is registered in the user profile 18 of the user A. Similarly, “memory patent” and “SRAM” are registered in the user profile 18 of the user B, and “semiconductor manufacturing equipment” is registered in the user profile 18 of the user C.

As described above, if the feedback to the common profile 10 and the feedback to the user profile 18 are separately performed, it is necessary to update the general general search conditions required over time, Customization can be done in one framework. In the case where a filtering result customized for each user 3 is presented to a plurality of users 3, the conventional system has only the profiles 18 for each user 3 and updates them individually. Then, since the common feedback information is reflected only in one common profile 10 and only the difference from the common profile 10 needs to be described in the user profile 18, it is more efficient from the viewpoint of the processing amount and the storage capacity. It is.

(Display of search results classified for common profile 10 and user profile 18)
In the present embodiment, the common profile 10 and the user profile 18 coexist, but it is also conceivable to present information on this to the user.

FIG. 21 shows an example of the common profile 10 and the user profiles 18 for the two users 3.
In the common profile 10, search conditions such as (condition 1), (condition 2), (condition 3),... And search terms corresponding to the search conditions are described. The user profile 18 for the user A includes Search conditions such as (condition A1), (condition A2), and (condition A3) and search words corresponding thereto are described. The search for the user A is performed using both of them. Suppose that three articles for user A are obtained. Of these, “Article 1” is an article that satisfies (condition 1) in FIG. 21, “Article 2” is an article that satisfies (condition 2) and (condition 3), and “article 3” is , (Condition A1) and (Condition A3), an article presentation method as shown in FIG. 22 can be considered.

As shown in the figure, the search results are classified into “articles that match the search conditions of the common profile” and “articles that match the search conditions of your personal profile”, and “article 1” and “article 2”. "Is presented in the former, and" Article 3 "is presented in the latter. Thereby, the user 3 can easily know whether the presented article is adapted to the interests common to the group or to the personal interests.

FIG. 23 is a modification of FIG. In this example, information on the contribution of the common profile 10 and the contribution of the user profile 18 is added to each article. For example, “Article 1” is given 30 partial points because it satisfies three of the search conditions of the common profile 10 and 70 points because it satisfies seven of the search conditions of the user profile 18. Information that a partial point has been given is displayed as shown. In this example, it can be seen that "article 1" is the one that most suits the personal interest, and "article 3" is the one that most suits the interest common to the group.

情報 Along with the search results as shown in FIGS. 22 and 23, information about the ratio of the size of the common profile 10 to the size of the user profile 18 may be provided. Here, the size of the profile refers to a value such as the number of search words and search conditions in the profile and the sum of the weights of the search words. For example, a user with a "word count of common profile: word count of user profile" of 50:20 may have other members of the group who reflect much more personal interests than a user of which it is 50: 3. Shows that different search results are obtained.

(Third embodiment)
Next, a usage form of the information filtering system according to the third embodiment will be described with reference to FIG.
This embodiment is the same as the first and second embodiments, except that there is no need to always have a plurality of users 3.

FIG. 25 shows a device configuration in the present embodiment.
The difference from the first and second embodiments is that a user information management unit 20 is provided between the user information analysis unit 12 and the user information storage unit 13. The functions of the text information analysis unit 16, the text information output unit 15, and the user information input unit 11 are the same as in the first embodiment. The function of the user information analysis unit 12 is the same as that shown in FIGS. 11, 12 and 14 except that the analyzed user information is passed to the user information management unit 20 instead of being directly stored in the user information storage unit 13. And different. Therefore, here, only the functions of the text information search unit 14 and the user information management unit 20 different from those of the first embodiment will be described.

FIG. 26 shows an example of the flow of processing of the text information search unit 14 in the present embodiment.
Here, a description will be given of an example of a search method in which a similarity between the profile 19 and the text information is calculated, and a search is performed by ranking the similarity. The text information search unit 14 performs the following processing on all newly arrived text information.

First, the similarity between the profile 19 and the text information is calculated according to the ordinary search method (step K3). Next, the current time is added to the search conditions in the profile 19 that match the text information in the above-described similarity calculation (step K4). Similarly, the current time is added to the search words in the profile 19 that match the text information in the above-described similarity calculation. In the present embodiment, these times added to the search condition and the search word are referred to as the latest matching time. When the above processing for all text information is completed (Y in step K6), the text information search unit 14 ranks the text information in order of similarity, and passes the result to the text information output unit 15 (step K7). The only difference between the function of the text information search unit 14 in the present embodiment and that in the first embodiment is that the former allows the current time to be entered in the search conditions and search words in the profile 19.

FIG. 27 shows an example of a search condition and a search word to which the latest matching time is added.
As specific search conditions and search terms, for example, those shown in FIG. 4 can be considered. Here, there are four search conditions, and (search word A) to (search word F) are specified in each condition. Then, the latest matching time is given to each search condition and search word. For example, if certain text information includes (search term A) and (search term B) and these search terms satisfy (search condition 1) at “12:25 on March 6,” 1), (search term A) and (search term B) each have a latest matching time of “12:25 on the 6th”. Next, assuming that another piece of text information includes (search term A), which satisfies (search condition 1) at “11:30 on September 9”, (search condition 1) and (search term B) Is added with the latest matching time of “11:30 at 9th”. As a result, the profile 19 becomes as shown in FIG. As described above, the latest matching time indicates when the search condition or search word has recently been matched. In other words, the fact that this time is old indicates that the search condition or the search word has not been used recently, and that this time is new means that the search condition or the search word is effectively used in the search even now. It shows that it is being done.

FIG. 28 illustrates an example of a processing flow of the user information management unit 20 in the present embodiment.
The user information management unit 20 first receives from the user information analysis unit 12 the profile 19 to which the latest matching time as shown in FIG. Next, the latest matching time added to each search condition and search word in the profile 19 is compared with the current time. Then, search conditions and search terms that do not match any text information for a certain period are deleted from the profile 19, and the profile thus modified is stored in the user information storage unit 13. For example, it is assumed that the current time is “12:30 on the 26th”, and that the user information management unit 20 deletes search conditions and search terms that do not match for 20 days or more. At this time, when the profile 19 as shown in FIG. 27 is passed to the user information management unit 20, the latest matching time that is not used for 20 days or more is “6:12:25” (search condition 1). (Search term A) and (search term C) of (search condition 2). Therefore, if these are deleted from the profile 19, an updated profile 19 as shown in FIG. 29 is obtained. Alternatively, as a modified example, when a weight is given to each search condition or search word in the profile 19, it is possible to reduce only the weight without deleting the old search word.

に よ り Through the above processing, new search terms are added to the profile by relevance feedback, while old search conditions and search terms that are no longer useful for search are deleted from the profile. Thus, it is considered that the profile can be kept up to date.

(Fourth embodiment)
Next, a fourth embodiment of the present invention will be described.
The usage mode and the device configuration of the information filtering system 1 in the present embodiment are the same as those in the third embodiment. The functions of the text information analysis unit 16, the text information search unit 14, the text information output unit 15, and the user information input unit 11 are the same as in the first embodiment. Therefore, only the functions of the user information analysis unit 12 and the user information management unit 20 will be described here.

FIG. 30 illustrates an example of a processing flow of the user information analysis unit 12 in the present embodiment.
This substantially corresponds to that shown in FIGS. 11, 12 and 14 in the first embodiment. However, when the profile 19 is first generated or the profile 19 is updated by the relevance feedback, each search condition The second embodiment differs from the first embodiment in that the current time is added to each search word. That is, in the initial selection information analysis process (Y in step M2), the time at which they were generated is added to each search condition and each search word in the profile 19 (step M5). In the analysis process (N in Step M2), the time is added to the search condition and the search word newly added by the relevance feedback method (Step M7). These times added to the search condition and the search word are referred to as profile registration times here. For example, in a certain profile 19, it is assumed that the time of (search condition 1) is "6:12:25" and only (search word A) is specified as a search word corresponding to this condition. Then, it is assumed that (search term B) is newly added to (search condition 1) by “relative feedback” at “11:30 on 9th”. At this time, the profile registration time of “9:30 11:30” is added to (search term B), and the profile registration time of (search condition 1) is also updated to “9th 11:30” at the same time. The new profile is as shown in FIG. As described above, the profile registration time indicates when the search condition or search word was registered in the profile.

FIG. 31 shows an example of the flow of processing of the user information management unit 20 in the present embodiment.
First, the user information management unit 20 receives the profile 19 to which the profile registration time is added as shown in FIG. 27 from the user information analysis unit 20 (step N1). Next, the current time is compared with the profile registration time added to each search condition and search word in the profile 19 (step N2). Then, the search condition and the search word which have been registered for the profile 19 for a certain period of time are deleted from the profile 19 (step N3), and the profile 19 thus modified is stored in the user information storage unit 13 (step N4). For example, it is assumed that the current time is “12:30 on the 26th”, and that the user information management unit 20 deletes the search condition and the search word 20 days or more after the registration. At this time, when the profile 19 as shown in FIG. 27 is passed to the user information management unit 20, the latest matching time that has been registered for 20 days or more is “6th 12:25” (search Only (search term A) of condition 1) and (search term C) of (search condition 2). Therefore, if these are deleted from the profile, an updated profile as shown in FIG. 29 is obtained. As in the case of the third embodiment, as a modified example, when each search condition or search word in the profile 19 is weighted, the old search condition and search word are deleted. Instead, it is also conceivable to reduce only the weight.

The difference between the third embodiment and the present embodiment is that the former adds the latest time at which the search condition and the search word matched the text to the profile at the time of search, whereas the latter adds the search condition and the search word to the text. The time registered in the profile is added at the time of relevance feedback. In the present embodiment, as in the third embodiment, new search terms are added to the profile by relevance feedback, while old search conditions and search terms that are no longer useful for search are deleted from the profile. Thus, it is considered that the profile can be kept up to date.

The information filtering device according to the present invention described in the first to fourth embodiments is not built only in a distributed network environment, but can also be built on a personal computer operating in a single environment. .

FIG. 2 is a conceptual diagram showing a use form of the information filtering system according to the first embodiment of the present invention. FIG. 1 is a diagram showing a device configuration of an information filtering system according to first and second embodiments of the present invention. FIG. 4 is a diagram illustrating an example of a processing flow of a text information analysis unit according to the first and second embodiments of the present invention. FIG. 7 is a diagram illustrating an example of text information represented by a text information analysis unit according to the first and second embodiments of the present invention. FIG. 4 is a diagram illustrating an example of a processing flow of a text information search unit according to the first embodiment of the present invention. FIG. 7 is a diagram illustrating an example of a common profile according to the first and second embodiments of the present invention. FIG. 4 is a diagram illustrating an example of a processing flow of a text information output unit according to the first embodiment of the present invention. FIG. 4 is a diagram illustrating an example of a processing flow of a user information input unit according to the first and second embodiments of the present invention. The figure which shows the example of the relevance feedback information in the 1st and 2nd embodiment of this invention. The figure which shows the example of the relevance feedback information in the 1st and 2nd embodiment of this invention. FIG. 5 is a diagram showing an example of a processing flow of a user information analysis unit in the first and second embodiments of the present invention. FIG. 4 is a diagram showing an example of the flow of an initial selection information analysis process in the first and second embodiments of the present invention. The figure which shows an example of the topic knowledge in the 1st and 2nd embodiment of this invention. The figure which shows an example of the flow of the relevance feedback information analysis processing in 1st Embodiment of this invention. FIG. 5 is a diagram illustrating an example of relevance feedback information obtained from a plurality of users according to the first embodiment of the present invention. The figure which shows the modification of the relevance feedback information obtained from several users in 1st and 2nd embodiment of this invention. FIG. 4 is a conceptual diagram showing a use form of the information filtering system according to the second embodiment of the present invention. FIG. 9 is a diagram illustrating an example of a processing flow of a text information search unit according to a second embodiment of the present invention. FIG. 9 is a diagram illustrating an example of a processing flow of a text information output unit according to a second embodiment of the present invention. The figure which shows an example of the flow of the relevance feedback information analysis processing in 2nd Embodiment of this invention. FIG. 10 is a diagram illustrating an example of a common profile and a user profile for two users according to the second embodiment of the present invention. The figure which shows an example of the presentation method of the article in the second embodiment of the present invention. The figure which shows the modification of the presentation method of the article in 2nd Embodiment of this invention. FIG. 7 is a conceptual diagram showing a use form of the information filtering system according to the third and fourth embodiments of the present invention. The figure which shows the device structure of the information filtering system concerning 3rd and 4th embodiment of this invention. The figure which shows an example of the flow of a process of the text information search part in 3rd Embodiment of this invention. FIG. 14 is a diagram illustrating an example of search conditions and search words to which the latest matching time / profile registration time is added according to the third and fourth embodiments of the present invention. The figure which shows an example of the flow of a process of the user information management part in 3rd Embodiment of this invention. The figure which shows an example of the profile updated based on the latest adaptation time in 3rd Embodiment of this invention. The figure which shows an example of the flow of a process of the user information analysis part in 4th Embodiment of this invention. The figure which shows an example of the flow of a process of the user information management part in 4th Embodiment of this invention.

Explanation of reference numerals

REFERENCE SIGNS LIST 1 information filtering system 2 text information source 3 user 10 shared profile 11 user information input unit 12 user information analysis unit 13 user information storage unit 14 text information search unit 15 ... text information output unit, 16 ... text information analysis unit, 17 ... text information storage unit, 18 ... user profile, 19 ... profile, 20 ... user information management unit, 101 ... analysis dictionary, 102 ... topic knowledge.

Claims (2)

In an information filtering device that selects desired text information from a plurality of text information and presents it to a user,
Means for retaining search conditions;
Means for selecting the text information that matches the search condition held by the means,
Means for presenting the text information selected by this means to a user;
Means for adding time information to a search condition or search word to which the text information presented by this means matches;
Means for correcting the search condition based on the time information added by this means. In an information filtering device that selects desired text information from a plurality of text information and presents it to a user,
Means for retaining search conditions;
Means for selecting the text information that matches the search condition held by the means,
Means for presenting the text information selected by this means to a user;
Means for collecting feedback information that is a user's evaluation result for the text information presented by the means,
First correction means for correcting the search condition based on the feedback information collected by this means,
Means for adding time information to a newly added search condition or search word at the time of correction by the first correction means;
An information filtering device comprising: a second correction unit configured to correct the search condition based on the time information added by the unit.

Images