JPH1091645A - Method for retrieving information - Google Patents

Abstract

PROBLEM TO BE SOLVED: To avoid a state that a word deeply related to a keyword but regarded as a generally unimportant word is not extracted as a word to be retrieved in the case of retrieving required information from a data base by using a keyword and a word related to the keyword or regarded as an improtant word. SOLUTION: The relative importance of a word included in the range of a fixed distance from an inputted keyword (S10) is calculated from the absolute importance of the word and the absolute importance of a word adjacent to the word concerned without simply using only the absolute importance of the word (S12) and whether the word is to be used for retrieval or not is determined by using the relative importance (S13).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for retrieving desired information from a database, and more particularly to a method for retrieving desired information not only based on a keyword for retrieving the desired information but also on a word related to the keyword. How to find information.

[0002]

2. Description of the Related Art Recently, as a method of searching for desired information from a database composed of documents and the like, it has been widely used to use a keyword for search and a word related to the keyword together. In this search method, a searcher first selects a keyword. Next, a word having a certain degree of importance or a word having a certain degree of closeness is extracted from a thesaurus file storing words, importance of the words themselves, degree of perspective between words, and the like. Then, based on the keyword and the extracted words, desired information is searched from a database.

[0003]

However, in such an information search method, the significance of words themselves, the degree of perspective between words, and the like stored in the thesaurus file are determined by the meaning of words, the frequency of use, and the number of words. Grammatical position of
The word is predetermined and fixed from the viewpoint of the linguistic meaning that the word originally has. Thus, for example, if a word is generally of low importance, but is strongly related to the selected keyword, it is once extracted from the thesaurus file because it is strongly related to the keyword. However, the word itself is not so important, and as a result, the word is not extracted. As a result, there has been a problem that the intention of the searcher cannot be sufficiently reflected in the word group on which the search is based.

[0004]

In order to solve the above-mentioned problems, in an information search method according to the present invention, first, a word used for searching desired information from a database, that is, a keyword is selected and input. I do. Next, from a thesaurus file that stores words, a perspective indicating whether words are near or far from each other, and an absolute importance indicating how important words are generally, based on perspective, , A word within a certain distance from the keyword is detected. Next, the relative importance, which indicates how important the detected word is in relation to the keyword,
It is calculated based on the absolute importance of the detected word itself and the absolute importance of words located around the word. Then, words whose relative importance exceeds a certain threshold are extracted from the detected words. Finally, using the extracted words and keywords, the database is searched for desired information.

As described above, in the information retrieval method according to the present invention, words to be used for retrieval are extracted using the relative importance of words. By such means, words that are generally less important but should be identified as important in relation to the selected keyword, or more precisely, in relation to words related to the selected keyword Can be extracted without leakage. As a result, it is possible to sufficiently reflect the searcher's intention of the search.

[0006]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An information retrieval method according to the present invention will be described.
A description will be given according to the embodiment. Specific examples 1, 2, and 3 are given as embodiments. The main characteristic point of the specific example 1 is that when a word related to an input keyword is extracted from a thesaurus file, the perspective is in a certain value range in relation to the keyword, and
To extract words whose relative importance exceeds a certain value. The main feature point of the specific example 2 in which the specific example 1 is more preferable is that weighting depending on the perspective is performed when calculating the relative importance. Similarly, the main characteristic point of the specific example 3 in which the specific example 1 is changed to a more desirable form is to refer to the relative importance in the past search when calculating the relative importance.

<Configuration of Specific Example 1> FIG. 1 is a search flowchart of the specific example 1, and FIG. 2 is a block diagram of an information search device for realizing the information search method of the specific example 1. First, the configuration will be described with reference to FIG. 2, and then the search procedure will be described with reference to FIG.

<Structure of Information Retrieval Apparatus> As shown in FIG. 2, the information retrieval apparatus comprises a display / operation unit 1 and a thesaurus.
It comprises a file unit 2, a database unit 3, and a control unit 4. The display operation unit 1 inputs a keyword, which is a word necessary for a searcher to search for desired information,
It has a function of displaying the results of the search.

The thesaurus file section 2 stores data such as a plurality of words, the absolute importance of words, and the degree of perspective between words. Here, the “word” is a term including, for example, a computer, an IC, a vacuum tube, a gear, and the like. The “absolute importance of a word” is the importance of the word calculated based on the meaning and appearance frequency of the word. For example, the absolute importance of the frequently used word “IC” is , The word "vacuum tube" which is used less frequently is greater than the absolute importance. Furthermore, "Perspective between words"
Is calculated based on the general meaning of each word,
The degree of association between words refers to, for example, words “computer” and word “IC” are close, while words “computer” and word “gear” are far apart (hereinafter, when the distance is short, “ "The degree of perspective is small," and when the distance is long, "the degree of perspective is large." Note that the degree of perspective between words is determined in specific fields such as electrical relations and chemical relations.
It is desirable to calculate based on the degree of co-occurrence of whether words are used simultaneously. For example, when the field is limited to "early computer", the word "computer" and the word "gear" appear frequently at the same time, that is, the co-occurrence degree becomes large. Becomes smaller.

The database section 3 stores data of documents such as academic journals and academic journals in a code format such as JIS code. The control unit 4, which plays a central role in information retrieval, includes a detection unit 4A, a calculation unit 4B, an extraction unit 4C, and a search unit 4D. The detection unit 4A includes:
A related word is detected from the thesaurus file unit 2 based on the input keyword. The calculation unit 4B calculates the relative importance of the detected word, as described later. The extracting unit 4C extracts words having a relative importance of a certain level or more from the detected words. The search unit 4D searches the database 3 for documents containing keywords and the extracted words.

<Structure of Thesaurus File> FIG.
FIG. 4 is a diagram illustrating a data structure of a thesaurus file unit.
In this figure, each node has a word (k1, k2,...)
…) And the absolute importance of the word (0.8, 0.5, ……)
And Further, the link represents a higher / lower relationship, synonym / synonymous relationship, degree of co-occurrence, etc. between words. Represents. Hereinafter, the word is represented by ki, the perspective of the word ki and the specific word K is represented by D (ki, K), and the relative importance of the word ki to the specific word K is represented by R (ki, K). Notation (However,
i is any integer). The thesaurus composed in this way
The file unit 2 extracts words that are located within a certain range and have a certain importance or more with respect to the input keyword based on the threshold of the perspective and the threshold of the relative importance.

<Operation of Specific Example> As shown in FIG. 1, the search procedure includes an input step, a detection step, a calculation step,
It consists of an extraction step, a search step, and an answer step. Hereinafter, those steps will be described in detail. In the following description, the input keyword is K, the perspective threshold is 2, and the relative importance threshold is 0.6.
And It is also assumed that K and k1 are the same word.

<Input Step> Step S10: The searcher inputs a keyword K expected to be necessary for obtaining a desired document from the display operation unit 1. Then, the input keyword K is sent to the detection unit 4A in the control unit 4.

<Detection Step> Step S11: The detection unit 4A detects a word having a perspective of 2 or less for the keyword K in the thesaurus file unit 2. More specifically, a word in which the number of links connected in cascade starting from the keyword K is in a range of two or less is detected. FIG. 4 is a diagram illustrating a range of the detected word. Here, if a set of detected words is represented by PR (K), PR (K) = ｛k1, k2,
k3, k4, k5, k6, k7, k8, k9, k10,
k11}. However, since k1 and K are the same word, k1 is described as K in the figure.

<Calculation Step> Step S12: The calculation unit 4B calculates the relative importance of each detected word. Here, the relative importance R
(Ki, K) is the absolute importance threshold 0.6 of the relative importance.
For the word ki larger than, the absolute importance is regarded as the relative importance. On the other hand, for the word ki whose absolute importance is smaller than the threshold 0.6 of the relative importance, FIG.
As shown in 1), when a set of words kj adjacent to a word ki and belonging to a set PR (K) is represented by A (ki), the absolute importance and relative importance of the words in the set A (ki) are represented. The relative importance is determined by adding the average value of the difference between the importance threshold 0.6 and the absolute importance of the word ki.
As shown in FIG. 12 (a2), if necessary,
By multiplying the reflection coefficient ε indicating how much the absolute importance of the adjacent word kj is reflected, the adjacent word kj
To increase or decrease the effect of the absolute importance of.

<Extraction Step> Step S13: The extraction unit 4C extracts words whose calculated relative importance is larger than the threshold value of the relative importance from the set PR (K). FIG. 5 is a diagram showing the range of the extracted words. Here, the set of extracted words is RW
Expressed as (K), RW (K) = ｛K, k2, k3,
k4, k5, k6, k8, k9, k10, k11}.

<Search Step> Step S14: The search unit 4D searches for the extracted word,
Using the logical OR of the keyword K input first, the database unit 3 searches for documents containing those words and keywords. <Answer Step> Step S15: The display operation unit 1 displays the content of the searched document to the searcher.

<Effect of Specific Example 1> As described above, in the information search method of the specific example 1, first, a keyword for searching for information is input. Next, a word having a certain degree of perspective with respect to the input keyword, that is, a word located within a certain distance from the keyword is detected. Then, the relative importance of the detected word with respect to the keyword is determined by using the absolute importance inherent in the word itself and the absolute importance inherent in one or more words adjacent to the word itself. And calculate. Further, from the detected words, only words whose calculated relative importance is larger than a predetermined threshold, that is, words that are more important than a certain level are extracted. Finally, using the keyword and the extracted word, a document or the like containing the keyword or the word is searched.

As described above, unlike the conventional information search method, at the word extraction stage of determining whether or not a word closely related to the input keyword is useful for search, Instead of using the absolute importance of the word itself, the relative importance calculated from the absolute importance of words adjacent to the word is used. This makes it possible to extract words that should not be extracted by the conventional information search method and should be recognized as important in relation to surrounding words. As a result, your search intent can be more clarified than with traditional information search methods,
Information can be searched with higher accuracy.

When the absolute importance of the detected word itself is larger than the threshold value, that is, when the word is always extremely important, the relative importance is not calculated again. The relative importance is calculated only when the absolute importance of the word itself is smaller than the threshold. This makes it possible to quickly determine whether the word is useful for a search.

When calculating the relative importance of a detected word, a reflection coefficient indicating how much the absolute importance of a word adjacent to the word is reflected is used. This makes it possible to arbitrarily adjust the magnitude of the influence of the absolute importance of an adjacent word on a certain detected word, so that the intention of the search can be made clearer.

The degree of perspective between words is determined based on the meaning of the words. This makes it possible to detect words that are close to the keyword in terms of the general meaning that the word originally has. Further, the degree of perspective between words is determined based on the degree of co-occurrence indicating whether or not words are present together in a specific field at the same time. As a result, it is possible to detect a word that is generally weakly connected to the selected keyword, but is strongly connected in a specific field.

<Structure of Specific Example 2> An information search method of specific example 2 will be described. The configuration of the information search device for realizing the specific example 2 is almost the same as the information search device of the specific example 1. The main difference is the configuration of the control unit 4. FIG.
FIG. 13 is a block diagram of a control unit in a specific example 2. The control unit 4 has a configuration in which a weighting unit 4E is added to the control unit 4 of the first embodiment.

<Operation of Embodiment 2> The search procedure of the information search method of Embodiment 2 is almost the same as the search procedure of the information search method of Embodiment 1. The main addition is that in the calculation step of calculating the relative importance, weighting depending on the perspective is performed. FIG. 7 is a search flowchart of the second embodiment. Hereinafter, the weighting step will be described in detail while following this flowchart. It is assumed that the perspective threshold and the relative importance threshold are the same as those in the first specific example (the perspective threshold is 2 and the relative importance threshold is 0.6).

<Input Step to Detection Step> Steps S20 to S21: The searcher inputs a keyword K, and the keyword K is passed to the detection unit 4A in the control unit 4. Next, the detection unit 4A determines that the keyword K has a perspective degree of 2 in the thesaurus file unit 2.
By detecting the following words, the set PR (K)
Generate

<Calculation Step> Step S22a: The calculation unit 4B calculates the relative importance R (ki, K) for each word in the set PR (K). Step S22b: The weighting unit 4E determines the relative importance R
(Ki, K) and a weighting coefficient δ that is a function of perspective.
(D (ki, K)) is used to calculate the weighted relative importance. Here, the weighted relative importance is Rp (ki,
Indicated by K), it is calculated by the equation shown in FIG. In particular, when the degree of weighting is to be reduced as the degree of perspective is larger, that is, as the distance from the keyword K is increased, the weighting coefficient δ which is a fixed value is set as shown in FIGS. And perspective D (k
i, K).
Note that here, according to the equation of FIG.
Assume = 0.1.

<Extraction Step> Step S23: The extraction unit 4C calculates the weighted relative importance Rp (k) from the words of the set PR (K).
i, K) are extracted for words whose relative importance is greater than the threshold value of 0.6. FIG. 8 is a diagram showing the range of the extracted words. Here, when a set of extracted words is represented by RpW (K), RpW (K) = ｛k2, k3, k4, k5,
k8, k9, k11}.

<Search Step to Answer Step> Steps S24 to S25: The search unit 4D searches the database unit 3 for a corresponding document using the extracted word and the keyword K input first. Then, the display operation unit 1 answers the content of the searched document.

<Effect of Specific Example 2> As described above, in the information search method of specific example 2, unlike the information search method of specific example 1, it is determined whether or not the detected word is useful for the search. In the extraction step of the word to be performed, the calculated relative importance is corrected based on the degree of perspective between the keyword and the word itself without using the calculated relative importance as it is. As a result, the relationship between whether the keyword and the word are close or far from each other in terms of the original meaning,
This can be reflected on the relative importance.

Further, the correction is made so that the calculated relative importance becomes smaller as the perspective becomes larger, that is, as the keyword is separated from the detected word. As a result, even though the keyword is distant from the input keyword, the relative importance increases because the absolute importance of the adjacent word is high, and as a result, the word is extracted as useful for search. Can be avoided.

<Structure of Specific Example 3> An information search method of specific example 3 will be described. The configuration of the information search device for implementing the third embodiment is also substantially the same as the configuration of the information search device of the first embodiment. The main difference is the configuration of the control unit 4. FIG. 9 is a block diagram of a characteristic portion in the information search device of the third embodiment. This information search device has a configuration in which a history reference unit 4F for referring to the history of past searches and a history storage unit 5 for storing the history of past searches are added to the information search device of Example 1. .

<Operation of Embodiment 3> The search procedure of the information search method of Embodiment 3 is almost the same as the search procedure of the information search method of Embodiment 1. The main addition is that in the calculation step of calculating the relative importance, the relative importance in the past search is referred to. FIG. 10 is a search flowchart of the third embodiment. Hereinafter, the calculation step will be described in detail while following this flowchart. In the past, searches S, T, and U with keywords Ks, Kt, and Ku were executed in that order, and a history of words detected in each of the searches and their relative importances is shown in FIG. It is assumed that the data is stored in the history storage unit 5 in the following manner.

<Input Step to Detection Step> Steps S30 to S31: The searcher who has completed the searches S, T, and U further inputs a keyword K, and
Is transferred to the detection unit 4A in the control unit 4. Next, the detecting unit 4A creates a set PR (K) by detecting words in the thesaurus file unit 2 having a perspective of 2 or less with respect to the keyword K.

<Calculation Step> Step S32a: The calculation unit 4B calculates the relative importance R (ki, K) for each word in the set PR (K). Step S32b: The history reference unit 4F calculates the calculated relative importance R (ki, K), the word ki detected in the past searches S, T, U stored in the history storage unit 5, and the word ki. Relative importance of ki R (ki, Ks), R (ki,
Kt) and R (ki, Ku) are used to calculate the relative importance with history.

Here, the relative importance with history is defined as Rh (k
i, K), as shown in FIG. 14 (c1),
It is calculated as the average value of the product of the relative importance R in each search and the history reflection coefficient α which is a function of the search time and indicates how much the search is reflected. More specifically, assuming that the keyword also represents time, for example, the relative importance Rh (ki, K) with history of the word ki when a search is performed at time K based on keyword K is ｛R (ki, K). )
* Α (K) + R (ki, Ks) * α (Ks) + R (k
i, Kt) * α (Kt) + R (ki, Ku) * α (K
u)｝ / 4.

It should be noted that, when the original relative importance R (ki, K) is larger than the threshold value of the relative importance, the expression (c1) holds the original relative importance R (ki, K) as it is. It is regarded as the relative importance Rh with history (ki, K).
On the other hand, when the original relative importance R (ki, K) is smaller than the relative importance threshold, the average value of the relative importance R (ki, Ks) in the past search is calculated, The relative importance with history Rh (ki, K) is set. Especially,
In order to assign more weight to the relative importance R as the search is performed at a point closer to the present time, as shown in FIGS. 14 (c2) and (c3), a history having a fixed value is used. It can be obtained using the reflection coefficient α and the search new / old degree t indicating how new or old the search time is (t is larger, the older the search time is).

<Extraction Step> Step S33: The extraction unit 4C selects, from the words of the set PR (K), the calculated relative importance Rh (k) with history.
i, K) are extracted for words whose relative importance is greater than the threshold value of 0.6. <Search Step to Answer Step> Steps S34 to S35: The search unit 4D searches the database unit 3 for a document using the extracted word and the keyword K. Then, the display operation unit 1 displays the content of the searched document.

<Effect of Specific Example 3> As described above, in the information search method of specific example 3, unlike the information search method of specific example 1, it is determined whether the detected word is useful for the search. In the word extraction stage, the relative importance calculated at the present time is corrected and used after simply referring to the relative importance in the past search without simply using the relative importance calculated at the present time. As a result, it is possible to reflect the relative importance based on the input keyword in the past series of searches, which is positioned as a footprint that makes the search intention clearer, so that the search intention can be made clearer It becomes possible. In addition, only when the relative importance at the present time is smaller than the threshold, the relative importance in the past search is referred to,
On the other hand, when the relative importance is large, the relative importance at the present time is used without reference to the relative importance in the past search. This makes it possible to quickly determine whether a word is useful for a search.

Further, when the search was performed,
That is, the relative importance calculated at the time of the search is weighted depending on whether the search is performed near or far from the current time. Thereby, the degree of referring to the relative importance in the search can be adjusted according to the search time, so that the intention of the search can be further clarified.

Furthermore, the closer the search time is to the current time, the greater the degree of referring to the relative importance calculated in the search. In general, the search intention becomes clearer as the time is closer to the present time, and thus, it is possible to extract words in which the search intention is more reflected.

[Brief description of the drawings]

FIG. 1 is a search flowchart of a first specific example.

FIG. 2 is a block diagram of an information search device of a specific example 1.

FIG. 3 is a diagram showing a data structure of a thesaurus file section.

FIG. 4 is a diagram showing a range of words detected in a specific example 1.

FIG. 5 is a diagram illustrating a range of words extracted in a specific example 1.

FIG. 6 is a block diagram of a control unit in a specific example 2.

FIG. 7 is a search flowchart of a specific example 2.

FIG. 8 is a diagram showing a range of words extracted in a specific example 2.

FIG. 9 is a block diagram of a characteristic portion of the information search device according to the third embodiment.

FIG. 10 is a search flowchart of a specific example 3.

FIG. 11 is a diagram illustrating a history stored in a history storage unit of a specific example 3.

FIG. 12 is a diagram (part 1) illustrating an expression used in a calculation step.

FIG. 13 is a diagram (part 2) illustrating an expression used in the calculation step.

FIG. 14 is a diagram (part 3) illustrating an expression used in the calculation step.

[Explanation of symbols]

 S10 input step S11 detection step S12 calculation step S13 extraction step S14 search step S15 answer step

Claims (11)

[Claims] 1. A step of inputting a keyword used for searching desired information from a database; a plurality of words; a degree of perspective between words; Detecting, from a file, words within a certain range of perspective from the keyword; and determining the relative importance of the detected word with respect to the keyword, the absolute importance of the word itself and the words adjacent to the word. Calculating based on the absolute importance, extracting a word whose relative importance is greater than a predetermined threshold from the detected words, and selecting a word from the database based on the keyword and the extracted word. A step of searching for information. 2. The calculating step according to claim 1, wherein the relative importance of the detected word with respect to the keyword is calculated based on the absolute importance of the word itself, the absolute importance of a word adjacent to the word, and the word. And a reflection coefficient indicating a degree of reflecting the absolute importance of a word adjacent to the word. 3. The calculating step according to claim 1, wherein, if the absolute importance of the detected word itself exceeds a predetermined threshold, the relative importance of the detected word with respect to the keyword is determined. It is calculated based on the absolute importance and the absolute importance of the word adjacent to the word. When the absolute importance of each detected word exceeds a predetermined threshold, the relative value of the detected word with respect to the keyword is calculated. An information retrieval method characterized in that importance is represented by the absolute importance of the detected word itself. 4. A step of inputting a keyword used for searching for desired information from a database; and determining a plurality of words, a perspective between words, and a thesaurus file storing the absolute importance of the words themselves. Detecting words in the range of perspectives of the words; determining the relative importance of the detected words with respect to the keyword; the absolute importance of the words themselves; the absolute importance of words adjacent to the words; Calculating based on relative importance weighting depending on the degree of perspective with the keyword; extracting words whose relative importance exceeds a predetermined threshold from the detected words; Searching for desired information from a database based on the words. 5. The information retrieval method according to claim 4, wherein the weight of the relative importance of the detected word is reduced as the detected word is farther from the keyword. 6. The method according to claim 1, wherein, between the calculating step and the extracting step, the calculated relative importance is corrected using a relative importance of the word with respect to a keyword in a search performed in the past. An information retrieval method, comprising the step of: 7. The information search method according to claim 6, wherein the correction is performed using weighting of relative importance depending on a point in time when the search is performed. 8. The information search method according to claim 7, wherein the closer to the present point in time of the search, the greater the weight of the relative importance of the word with respect to the keyword in the search. Method. 9. The method according to claim 6, wherein, if the absolute importance of the detected word does not exceed a predetermined threshold, the relative importance of the word with respect to the keyword is calculated.
Correcting using the relative importance of the word itself in the past, and when the relative importance of each detected word exceeds a predetermined threshold, the relative importance of the word to the keyword is used as it is. Information retrieval method. 10. The information retrieval method according to claim 1, wherein the degree of perspective between words is determined based on a normal meaning inherent to the word. 11. The information retrieval method according to claim 10, wherein the degree of perspective between words is determined based on the degree of co-occurrence indicating the degree to which words occur simultaneously in a specific field.