JPH05158991A - Information retrieval system - Google Patents

Abstract

PURPOSE:To provide the information retrieval system capable of giving the order of priority and specifying the retrieval not including a specific keyword. CONSTITUTION:In retrieval, keywords and documents are represented by nodes using a keyword network part 12 representing the relation between keywords, a retrieval keyword importance setting part 11 specifying the importance of the retrieval keyword, and an information storage part 13 storing information, and the relation of keywords is represented by the weight of the link between nodes to propagate the activity of nodes. Further, the retrieval keyword importance setting part 11 is provided with a retrieval history storage part 10 storing the retrieval condition and the degree of satisfaction of a retriever against the retrieval result under the retrieval condition. Based on the past retrieval history of the retriever stored in the retrieval history storage part 10, the retrieval condition is set while having a conversation with the retriever.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an information retrieval system using a neural network model in information retrieval such as document retrieval or data retrieval.

[0002]

2. Description of the Related Art A search method that uses a logical expression of a keyword as a search means cannot perform a search similar to human association, so a search method that expresses the relationship between keywords by a neural network has been proposed. It was This is disclosed in, for example, Japanese Patent Laid-Open No. 224068/1990, and is configured as shown in FIG.

In the prior art, a keyword correlation table created by previously calculating the correlation of keywords is created,
At the time of search, a keyword to be searched is converted using a keyword correlation table to calculate a search feature vector, and the degree of similarity between this and the stored data feature vector is obtained by a neural network. For example, document P1 is keyword A,
Suppose we have B. At this time, if there is a keyword correlation table as shown in Table 1, the contents of rows A and B of that table are referenced.

[0004]

[Table 1]

A = (1 0.1 0 0 0.1 0.9 0.8) B = (0.2 1 0 0.4 0.3 0.3 0) The feature vector of the document P1 is generated by averaging A and B. However, when one of the elements is 1, it is set to 1 without taking the average. As a result, the feature vector of P1 becomes P1 = (1 1 0 0.2 0.2 0.6 0.4). Next, when searching a document having the keyword A, the search feature vector q is q = (1 0.1 0 0 0.1 0.9 0.8) by referring to the keyword correlation table in FIG. A search is performed by inputting this search feature vector into the neural network of FIG.

[0006]

As described above, in the prior art, since there is no means for designating the importance of the search keyword, it is not possible to give the priority according to the search request of the searcher, and the search is performed. When there is more than one piece of information that has been retrieved, there is a problem in that all the retrieved information must be examined to find the optimum information. For example, when searching for information about a workstation, searchers who are primarily interested in its performance are primarily interested in workstation speed and memory size, and secondary in terms of price. For searchers, "the fastest in the world"
Such information is important. However, searchers who are interested in the sales and sales of workstations are mainly interested in the price, sales system, sales record, etc., and for such searchers, information such as "This workstation aims to sell XX units this year" is important. For example, keyword A
If "is price", keyword B is "speed", and keyword F is "sales record", the searchers who are mainly interested in the keyword "speed" and little interested in the keyword "price" are mostly interested. No sales record
Documents corresponding to such keywords will also be searched. Further, this conventional technique has a problem that it is not possible to specify that a search that does not include a specific keyword is specified.

The present invention has been made in order to solve the above-mentioned drawbacks of the conventional ones, and it is possible to give a priority according to a search request of a searcher, and further, a specific keyword is not included. It is an object of the present invention to provide an information retrieval system that can specify to retrieve things.

[0008]

An information retrieval system according to the present invention is an information retrieval system for retrieving data to which a keyword is added, and a keyword network section showing a relation between the keywords and a retrieval keyword of a retrieval person. Using the search keyword importance setting section that specifies the degree of importance of the keyword and the information storage section that stores information, the keywords and documents are expressed as nodes, and the relationship between the keywords is expressed by the weight of the link between the nodes. Then, the search is performed by propagating the activity of the node.

Further, in setting the search keyword importance, a search history storage unit for storing the search condition and the satisfaction of the searcher for the search result under the condition is provided in the search keyword importance setting unit, and the search history storage unit is provided. Based on the past search history of the searcher stored in, the search condition is set while interacting with the searcher.

[0010]

According to the present invention, a searcher can specify the importance of a plurality of search keywords, and information is searched with priorities. Also, a flexible search function can be realized according to the searcher's search intention. In addition, the priority of the retrieved information can be determined according to the degree of activity of the node. In addition, by activating the node corresponding to the search keyword with a negative value, it is possible to specify that a search is performed that does not include a specific keyword.

Further, if a search history storage unit is provided, keywords (plural) specified by the searcher in past searches are provided.
And the set values of their importance (these are referred to as weighted search conditions below) and the satisfaction of the searcher for the search results under those conditions are stored, and the conditions specified by the searcher in the current search are stored. The searcher sets the weighted search condition interactively by selecting the one close to the above from the weighted search conditions that the searcher showed high satisfaction in the past search, and the system presents it to the searcher. Since it can be done, efficient search is possible.

[0012]

【Example】

Example 1. Hereinafter, the details of the present invention will be described with reference to a document search to which a keyword is added as an embodiment. FIG. 1 shows the configuration of an embodiment of the present invention. The search keyword setting unit 11 that specifies the importance of the search keyword of the searcher includes a node Ui corresponding to the search keyword. In the keyword network unit 12 that indicates the relationship between keywords, the keyword Kj is connected to another keyword by a link whose weight that indicates the degree of association is Vij. Node Ui corresponding to a certain keyword and node Kj corresponding to the same keyword
Are linked by links of weight Wij. Information storage unit 13
Stores information and includes a node Dm corresponding to each information.
The node Kj on the keyword network corresponding to the keyword included in the document Dm is linked with the link having the weight Xjm. The search history storage unit 10 is not included in the first embodiment.

Next, the operation of this embodiment will be described with reference to FIG. First, in step 1, the search keyword node of the searcher
Depending on the importance of the searcher, Ui is set to, for example, the activity of the keyword of interest is 1.0, the activity of the keyword of little interest is 0.5, and the activity of the keyword of no interest is 0. Then, activate the keyword with a negative value if the keyword has a problem. There are several possible methods for setting the degree of activation, but the simple method is for the searcher to explicitly set the importance of each keyword. In step 2, the activity of the node Ui is propagated to the node Kj via the link Wij. (Activity of Kj) = Σ (Activity of Ui) × Wij However, Σ is performed for all nodes Uj connected to Ki. In step 3, the activity of node Kj is Ki through Vji.
Propagate to. Various methods can be considered for this propagation, for example, as follows. (Activity of Ki) = Σρ × (Activity of Kj) ÷ n × Vji However, Σ is performed for all nodes Kj connected to Ki. ρ is a damping constant (0 <ρ <1) n is the number of links connected to Kj In step 4, the activity of the node Kj is propagated to the node Dm through the link Xim. (Activity of Dm) = ΣKj × Xjm However, Σ is performed for all nodes Kj connected to Dm. In step 5, the degrees of activity of the nodes Dm corresponding to the respective documents are rearranged in order, and a document having a high degree of activity is set as a document having a high search priority. In this way, the searcher can specify the importance of multiple search keywords, and the information is searched with priorities, so the searcher's A flexible search function can be realized according to the search intention. That is, the priority of the retrieved information can be determined according to the degree of activity of the node. In addition, by activating the node corresponding to the search keyword with a negative value, it is possible to specify that a search is performed that does not include a specific keyword.

Example 2. Next, the invention described in claim 2 will be described. In this example, as shown in FIG. 1, the search keyword importance setting unit 11 includes a search condition and a search history storage unit 10 that stores the searcher's satisfaction with the search results under the search conditions, and searches in the past. , The keywords specified by the searcher, set values of their importance (these are referred to as weighted search conditions below), and the satisfaction of the searcher for the search results under those conditions are stored. , By selecting from the weighted search conditions that the searcher showed a high degree of satisfaction in the past search, the conditions close to the condition specified by the searcher in the current search, and the system presents them to the searcher to make the dialog Specifically, the searcher sets the weighted search conditions.

The operation will be described with reference to FIG. Steps 10, 11, and 12 correspond to step 1 in FIG. First, in step 10, the searcher inputs search keywords (plurality), and then the same as the search keywords (plurality) entered in step 11, or if there is no same, past search conditions similar to those keywords. From the search history storage unit, and presents the searcher with the weighted search condition found and the degree of satisfaction with the search result under that condition. In step 12, the searcher refers to the presented search history (weighted search condition and satisfaction) to add or delete the keyword in this search, and the node corresponding to the keyword obtained as a result is important to the searcher. Activate depending on the degree. The following steps 2, 3, 4, and 5 are the same as those described with reference to FIG. After step 5, in step 6, the satisfaction level of the searcher with respect to the search result is input and stored in the search history storage unit in combination with the weighted search condition.

[0016]

As described above, according to the present invention, there is provided a keyword network section showing a relation between keywords,
Using the search keyword importance setting unit that specifies the importance of the search keyword of the searcher and the information storage unit that stores information, the keywords and documents are expressed as nodes and the relationship between the keywords is expressed between the nodes. Since the search is performed by expressing the weight of the link and propagating the activity of the node, it is possible to set the priority according to the search request of the searcher, and further search for the one that does not include a certain keyword. You can specify that.

Further, in setting the search keyword importance, a search history storage unit for storing the search condition and the satisfaction of the searcher for the search result under the condition is provided in the search keyword importance setting unit, and the search history storage unit is provided. Based on the past search history of the searcher stored in, the search condition can be set while interacting with the searcher, thereby enabling an efficient search.

[Brief description of drawings]

FIG. 1 is an explanatory diagram showing a system configuration of an embodiment of the present invention.

FIG. 2 is a flowchart explaining the operation of an embodiment of the present invention.

FIG. 3 is a flowchart illustrating the operation of another embodiment of the present invention.

FIG. 4 is an explanatory diagram illustrating a conventional information search method.

[Explanation of symbols]

 10 search history storage unit 11 search keyword importance setting unit 12 keyword network unit 13 information storage unit

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Ryuichi Kobune 8-1-1 Tsukaguchi Honcho, Amagasaki City Mitsubishi Electric Corporation Central Research Institute (72) Inventor Takashi Yokota 8-1-1 Tsukaguchi Honmachi, Amagasaki Mitsubishi Electric Central Research Institute Co., Ltd. (72) Inventor Hiroki Konaka 8-1-1 Tsukaguchi Honcho, Amagasaki City Mitsubishi Electric Corporation Central Research Institute (72) Inventor Kazuhiro Abe 8-1-1 Tsukaguchi Honmachi, Amagasaki Mitsubishi Electric Central Research Institute Co., Ltd.

Claims (2)

[Claims] 1. An information retrieval system for retrieving data to which a keyword has been added, comprising: a keyword network section that represents the relationship between the keywords; and a retrieval keyword importance setting section that specifies the degree of importance of the retrieval keyword of the searcher. Using the information storage unit that stores information, the keywords and documents are expressed as nodes, and the relation between the keywords is expressed by the weight of the link between the nodes to propagate the activity of the node to perform a search. An information retrieval system characterized by performing. 2. In setting search keyword importance,
The search keyword importance setting unit includes a search history storage unit that stores the search condition and the satisfaction of the searcher for the search result under the condition, and stores the search history of the searcher stored in the search history storage unit. The search condition is set based on the dialog with the searcher.
Information retrieval system described in section.