JP2001256251A - Device and system for automatically evaluating document information - Google Patents

Abstract

PROBLEM TO BE SOLVED: To solve a problem such that it is necessary to manually prepare a keyword for previously specifying the destination of distribution when sorting and distributing the claim information of electronic mail or the like to appropriate persons in charge. SOLUTION: The electronic mail is regarded as a natural language text and the same method can be used even for inputting the inquiry or claim of audio or fax except for electronic mail. Information provided in a natural language is automatically evaluated and sorted by estimating the destination of distribution suitable for sorting from contents thereof. For example, it is estimated and sorted who is suitable for receiving the claim dispatched to the claim processing part of a manufacturer. When information to be inputted is expressed in the natural language, a format is no object. A title or keyword is not limited. The estimation rule of category can by preset but can be previously learnt by sample data or learnt after repeated operation as well.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method of analyzing the contents of an e-mail when receiving the e-mail or the like and distributing the contents to an optimal person in charge. The present invention relates to a document information automatic evaluation device and a document information automatic evaluation system for collating contents of a received e-mail, distributing the contents to a destination having a high degree of similarity, and appropriately correcting a word for determining the distribution destination.

[0002]

2. Description of the Related Art Inquiries about the functions and operation methods of products and services delivered by manufacturers and complaints about the products and services themselves are sent to the manufacturers on a daily basis. That is.
Manufacturers need to fully analyze these responses received from users, respond to those that should respond to users, and refer to parts that can be used for the development of the next product or service. For that purpose, it is necessary first to correctly assign the voices of these users promptly and appropriately to the appropriate person in charge.

[0003] On the other hand, if the number of reactions from the user is on the order of hundreds or thousands per day, enormous man-hours will be required only for the work of allocating to the appropriate person on the manufacturer side. There is a problem that it is difficult. With the development of the Internet and the fact that users can easily access manufacturers by means such as e-mail, this tendency is likely to become even more pronounced in the future.
In addition, not only manufacturers, but also problems with services of each municipality are often received from residents.
It seems that it is held by society in general.

For example, Japanese Patent Application Laid-Open No. 10-4428 discloses a keyword extracting means for analyzing the title and syntax of an e-mail and extracting a keyword from the e-mail when the e-mail is received. By providing a keyword setting means for setting a keyword in advance in a target mailbox, and a sorting means for comparing the keywords by the two means and transmitting the mail to a mailbox in which the keyword matches, the electronic mail is provided. The contents to be automatically sorted are described.

[0005]

The present invention has a content for quickly and automatically processing a large amount of received mail, as in the case of the above-mentioned electronic mail system.
However, in the above e-mail system, it is necessary to set a keyword in advance in a mailbox to be sorted. In addition, there is no description on how to handle the received e-mail if it is transferred to the wrong mailbox.

In the present invention, at the same time as addressing these problems, an e-mail is regarded as a natural language text, and the same method is applied to voices other than the e-mail (becoming text after speech recognition) or fax (becoming text after character recognition). ) Can be used for inquiries and complaints. It is a method of estimating a category suitable for classifying information provided in a natural language (free syntax expression in Japanese) from its contents, assigning an index, and automatically performing evaluation classification. For example, it is possible to analogize (categorical analogy) and classify a person who is suitable for receiving a complaint that has arrived at the complaint processing section of the manufacturer. The input information may be in any format as long as it is expressed in a natural language. There are no restrictions on titles or keywords. The category analogy rule can be set in advance, but it is also possible to learn in advance by using sample data, or to learn while repeating operations. Since the conventional sorting system mainly uses a method in which a keyword or the like is set in advance as in the above-described electronic mail system, this is considered to be a major feature of the present invention.

[0007]

According to a first aspect of the present invention, there is provided a document information automatic evaluation apparatus which uses pre-processing for generating an analogy rule for classifying documents and uses the analogy rule generated in the pre-processing. And a classification process for classifying and distributing claim data including claims and inquiries for each person in charge. The pre-processing is analogous to storing claim data for all distribution destinations whose distribution destinations are already known. A rule generation sample data file, an analogy rule generation unit that inputs the analogy rule generation sample data file and generates analogy rules composed of words or word strings for all distribution destinations, and the analogy rule generation unit that generates the analogous rule. An analogy rule file storing analogy rules, wherein the classification process inputs the claim data and converts it into a text format. An input unit, and a text file for storing the converted claims data to the text format,
A natural language analysis unit for inputting the claim data converted to the text file, performing morphological analysis and outputting as constituent word data composed of decomposed words, a constituent word data file storing the constituent word data, and the constituent words A classification unit that calculates the similarity by applying the analogy rule of the analogy rule file to each distribution destination for the constituent word data input from the data file, and distributes the claim data to the distribution destination having the highest similarity, Is provided.

According to the second aspect of the present invention, in the document information automatic evaluation device according to the first aspect, at the time of generation of the analogy rule in the preprocessing, the distribution destinations are already known for all the distribution destinations, and the distribution destinations are classified in advance. By inputting the complaint data provided, analogy rules for determining a distribution destination are generated for all distribution destinations.

A third aspect of the present invention provides the automatic document information evaluation apparatus according to the first aspect, wherein the claim data includes any of electronic mail, telephone voice information, voice mail, and fax information.

According to a fourth aspect of the present invention, in the first or second aspect, when the analogy rule is generated in the preprocessing, the distribution destinations are already determined for all the distribution destinations, and When the claim data classified in advance is input, the claim data is classified into two: claim data distributed to a certain destination and claim data distributed to a destination other than the destination, and is included in the two classified claim data. Find the ratio of the appearance frequency of each word or word string for all the words or word strings of the claim data of the delivery destination,
It is provided with a step of comparing it with a predetermined threshold value to generate an analogy rule for the distribution destination, and performing this for all distribution destinations.

According to a fifth aspect of the present invention, in the document information automatic evaluation device according to the first aspect, in the classification processing, the claim data is converted into text format data in the classification process, and further divided into word strings by morphological analysis. When the word data is compared with the analogy rule of each distribution destination for all the distribution destinations, the similarity is calculated and the maximum value among the calculated values is equal to or greater than a predetermined threshold value. Distributing the claim data to the distribution destination that has calculated the maximum value, and not distributing the claim data if the value is less than the threshold value.

[0012] In a sixth aspect of the present invention, in the automatic document information evaluation apparatus according to the first aspect, the analogy rule generation sample data file includes claim data or constituent word data corresponding to the claim data.

The document information automatic evaluation device according to a seventh aspect of the present invention, in the first aspect, wherein in the distribution of the claim data, the claim data which is not distributed below the threshold value or which is distributed to an unauthorized distribution destination. Claim data and the like are additionally registered in the analogy rule generation sample data file as sample data on the distribution destination that should be originally distributed, and learning is performed in such a manner that the analogy rule generation unit is activated and the analogy rule is reviewed. It is prepared.

According to an eighth aspect of the present invention, in the first or seventh aspect, the document information automatic evaluation device automatically starts the input section when claim data is received, and is not distributed below the threshold value. In the event of complaint data or claim data delivered to an unauthorized delivery destination, the analogy rule generation sample data file is automatically added to the analogy rule generation sample data file as sample data of the delivery destination to be originally delivered, and It is provided to automatically start up and automatically review the analogy rules.

An automatic document information evaluation system according to a ninth aspect of the present invention is a system for automatically distributing complaint data received by e-mail, telephone, voice mail, facsimile, etc. to a person in charge, and the destination of distribution is already known. Input claim data, find analogy rules consisting of words or word strings for all distribution destinations, convert newly received claim data to text format, perform morphological analysis, and construct word data consisting of decomposed words And calculating the similarity by applying the analogy rule for each of the distribution destinations to the constituent word data, and distributing the similarity to the distribution destination having the maximum value of the similarity.

A document information automatic evaluation system according to a tenth aspect of the present invention is the ninth aspect, wherein the HTML file in which the topic requested to be obtained is described and the HTML file in which the topic requested to be obtained is not described are sampled. Generates analogy rules as HTML and retrieves the latest HT from an HTML file search system such as a Web robot on the Internet.
The method includes obtaining an ML file and searching for an HTML file having a topic desired to be obtained based on the analogy rules.

[0017]

Next, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram of a document information automatic evaluation device of the present invention using a general-purpose computer or a personal computer. Claim data and inquiries (hereinafter, referred to as e-mail, voice mail, telephone, fax, etc.) arrived.
This is called claim data. ) Is classified according to the person in charge of processing the claim data. In the “pre-processing” on the left side of FIG. 1, the analogy rules applied when classifying the claim data are set in advance for each person in charge. The right side shows the configuration in the "classification process" in which the claim data is sorted according to the analogy rules created by the left side configuration. Although a voice recognition device and a character recognition device for arriving claim data are omitted in the drawing, it is assumed that voice and fax data are output as text data by these devices.

In the “pre-processing” configuration, an analogy rule generation sample data file 105 which is sample data of claim information determined to which person to distribute,
An analogy rule generation unit 10 that generates analogy rules based on the rule.
1 and an analogy rule file 106 that stores the generated analogy rules. The analogy rule generation sample data file 105 is composed of text data or constituent word data corresponding to the claim. In the case of constituent word data, the processing of the natural language analysis unit 103 not described in the analogy rule generation unit 101 is performed. It becomes unnecessary.

In the structure of the "classification process", the input unit 102 for inputting the claim data 107 arriving in the form of e-mail or the like, converting it into a text format, and outputting it to a text file 108, and inputting the text file 108 A natural language analysis unit 103 for decomposing into words including nouns and verbs and outputting the result to a constituent word data file 109; a constituent word data file 109;
6 includes a classification unit 104 that determines the person in charge of the distribution destination of the claim data 107, and a classification result 110.

The analogy rule generation unit 101 receives consultations and inquiries regarding complaints and products including complaints from users regarding the use of the products and the like by e-mail, voice mail,
An analogy rule file 1 for receiving the complaint data 107 by telephone, facsimile, etc., and classifying the complaint data 107 by person (category) who performs the corresponding process.
06 is generated in advance. The analogy rule is output for each person in charge of the distribution destination in the following manner.

First, sample data 1 for analogy rule generation
Prepare 05. For each person in charge of processing a complaint, the number of sample data to which the distribution destination is determined in the complaint data 107 (for example, 50 to 100 per person in charge)
The analogy rule is generated by performing the same processing as that of the natural language analysis unit 103 incorporated in the processing of the analogy rule generation unit 101 and extracting constituent words.
When the constituent word data is given as the sample data, the above-described processing becomes unnecessary. Hereinafter, the sample data of the person in charge used to generate the analogy rule will be referred to as positive example data, and the sample data of other persons will be referred to as negative example data. The positive example data of the person in charge A becomes negative example data from the person in charge B.

Next, characteristic words frequently used are extracted from the positive example data. Word selection and extraction are performed for all constituent words obtained by performing the natural language analysis. With respect to the word extraction criterion, the following values are obtained for all the words constituting the positive example data, and words whose values satisfy the respective selection criteria are extracted.

[0024]

(Equation 1)

Here, a: the number of words X included in the positive example data b: the number of words other than X included in the positive example data d: the number of words X included in the negative example data Number.

FIG. 4 shows the contents of the words included in certain positive example data. In the case where the words included in the positive example data for a certain distribution destination are composed of A, B. The respective appearance frequencies are shown. here,
Recall and precision have the following meanings, respectively. Recall: The percentage of the word used in the positive example data of all positive and negative example data (use frequency bias)
Is a measure of the uniqueness of a word,
The larger the value, the higher the uniqueness. Relevance: An evaluation criterion for the frequency of occurrence of words in positive example data is used such that even if the recall is high, the relevance is too low to meet the extraction criterion.

The word extraction criterion is targeted at one that satisfies the following two conditions. Recall> threshold of recall Precision> (average of relevance for all words in the positive example data-threshold of difference from average of relevance) To explain a specific example, it is assumed that "word A" exists only at one place in certain positive example data and does not exist in negative example data. At this time, "word A"
Has a reproducibility of 1, which exceeds the reproducibility threshold and is a candidate for extraction, but there is a problem in selecting rarely used words that are used only at one place. At this time, if it is assumed that the average value of the precision is 0.03 and the threshold value of the difference from the average value of the precision is 0.01, the words having the precision of> 0.02 are to be selected. If the amount of words constituting is not too small, the "word A" will not be a target for extraction.

As described above, the recall rate of each word,
Check the precision against the above two criteria and extract words.
Go on. Using the words extracted in “extracting words”, a rule for inferring claims to be delivered to each person in charge is generated. The following information is output as the analogy rule. When words extracted as analogy rules (word information) appear continuously in positive example data, a word string obtained by combining the words is used as an analogy rule. If not consecutive, a single word is used as an analogy rule.

Next, a weight is set for each of the selected analogy rules.

The recall rate of the analogy rule (uniqueness of the analogy rule) is obtained by the following equation, and is used as the weight of the analogy rule.

[0031]

(Equation 2)

Here, a: the number of analogy rules x included in the positive example data d: the number of analogous rules x included in the negative example data The data 107 can be distributed. The determination of the person in charge of processing the claim data 107 is started.
If the input medium of the claim data 107 is voice or fax, voice or text recognition processing is performed, and the claim data 107 is output to the text file 108 as text data.
We accept electronic text such as e-mail as it is,
Output to the text file 108 as text data.

The text file 108 output from the input unit 102 is input, subjected to morphological analysis by the natural language analysis unit 103, to list constituent words, and output to the constituent word data file 109. At this time, only the noun including the proper noun and the verb are extracted, and the verb unifies the inflected form.

The similarity between the constituent words and the existing category name in the analogy inference rule file 106 is calculated, and the category is assigned to the category having the highest similarity. Various methods are conceivable for calculating the similarity. For example, the similarity calculation is performed as follows using an analogy rule file created by the analogy rule generation unit 101. (1) A word or word string in the analogy rule is extracted from the text. If the word string pattern stored in the analogy rule exists in the text, the word string pattern is extracted. (2) Calculate the similarity. For example, let the value of the following equation be the similarity.

[0035]

(Equation 3)

Here, k: coefficient n: number of words or word string patterns of the analogy rule existing in the text Ri: weight of the analogy rule Ni: number of words constituting the analogy rule W: existing in the text data Total number of words (3) Determination of the person in charge of the destination The similarity is calculated for all categories, and if the largest value exceeds the threshold, the text data is classified into that category. The original claim data 105 is distributed to the person in charge as the classification result 110. If the largest value does not exceed the threshold, it will be delivered to the person in charge of the complaint that could not be classified.

The distribution destination is determined as described above. If the result of the similarity calculation for a certain claim data 107 does not exceed the threshold value or is not distributed to the correct distribution destination, The complaint data 107 is confirmed by the human eyes and distributed to the correct distribution destination.
In that case, the corresponding claim data is added as a new analogy rule generation sample data file 105 to the existing sample data of the correct distribution destination. Analogue rule generation sample data file 1 for other distribution destinations
In step 05, the analog inference rule generation unit 101 is activated as it is, and the analog inference rule file 106 is updated. The analogy rule of the distribution destination to which the sample data has been added may naturally be updated by the addition of positive example data, but the analogy rule of other distribution destinations will also have negative example data. May also be updated. In this way, the analogy rules grow to be better.

Further, from the viewpoint of automation of the above-mentioned processing, the input unit 102 is automatically activated when a new claim data 107 is received.
Claim data 107 that could not be delivered correctly
Automatically adds new sample data to the analogy rule generation sample data file 105 every time such claim data is generated, every time a certain number of cases are accumulated, or each time a certain period elapses.
If it is started, automation including a learning function becomes possible.

Next, the operation of the present invention will be described based on specific document examples.

The automatic document information evaluation apparatus and the automatic document information evaluation system of the present invention include the following two processes. The analogy rule generation unit 101 corresponding to “preprocessing” creates analogy rules for distributing claims to each person in charge. The analogy rule creation is performed by preparing a plurality of analogy rule generation sample data files 105 of claims for which a predetermined person is to be distributed to each person, approximately the same number, and performing the processing described above. Create analogy rules for each person in charge.

For example, suppose that an attempt is made to generate an analogy rule for the person in charge 1. Claim data 1 shown in FIG.
One example of 07 is sample data that has been decided to be distributed to the person in charge 1. The words "analog", "rule", "claim", and "person in charge" are extracted from the sample data, and the analogical rule of FIG. Suppose it was done.
(In the example of the analogy rule in FIG. 2B, the weight of the analogy rule is a value separately set according to the above-described processing.) The “pre-processing” is completed.

Next, claim data 10 actually arrived
7 can be started.
Claims and questions received by telephone, fax, mail, etc. are converted into a text file in the input unit 102.
At this time, the claims received by telephone are subjected to voice recognition processing, and the claims received by fax are subjected to character recognition processing. In addition, data unnecessary for classification, such as the header of an email, is removed.

Next, the natural language analysis unit 103 performs a morphological analysis process on the text file 108 to extract nouns and verbs constituting the text file 108, and generates a constituent word data file 109. At this time, the verbs unify the inflected forms. For example, when the claim of the claim data 105 illustrated in FIG. 3A arrives, the constituent word data illustrated in FIG. 3B is output. Next, the classification unit 104 compares the analogy rule file 106 of each person in charge created by the analogy rule generation unit 101 and calculates the degree of similarity with each analogous rule. When examining the similarity with the person in charge 1 of this claim, the analogy rule created for person 1 in FIG. 2B is used to determine the similarity of the constituent word data in FIG. calculate. At this time, the rule (word or word string) included in the analogy rule of FIG. 2B is a word for calculating the similarity. With respect to the words extracted here, the number of appearance frequencies in FIG. 3B and FIG.
The similarity is calculated using the similarity calculation formula described above, using the weight of the analogy rule in the analogy rule of (1). FIG. 3C shows the extracted words.

For the claim of FIG. 3A, the similarity is calculated based on the analogy rules of all persons in charge, and if the largest value is larger than a separately determined threshold, the similarity is calculated. The complaint is delivered to the person in charge who has the analogy rule that calculated the degree. If the maximum similarity is less than the threshold, the unsuccessful classification is distributed to the person in charge of processing the complaint.

The analogy inference rules are reconstructed as appropriate at some timing (for example, when the calculated similarity is low, but an important claim is received) or when it is distributed to the wrong person in charge. That is, a rule for analogy can be learned. For example, even if the distribution destination is wrong, the claim data 107 is returned to the person in charge of processing the above-mentioned uncategorized complaint. In this case, the distribution destination is determined by human eyes. Also in this case, it is possible to add this data as sample data of the person in charge to be distributed, and to review the analogy rules again by the analogy rule generation unit 101. As described above, the claim data 107 whose distribution destination cannot be determined below the threshold value or the claim data 107
As for, it is necessary to select a correct distribution destination by human eyes, but such claim data 107 can be added as an analogy rule generation sample data file 105 for the correct distribution destination, and the analogy rule can be reviewed again. Wear. Further, it is possible to automatically execute this processing and to always keep a better analogy rule.

Next, a second embodiment of the present invention will be described.

HTML of a hypertext system provided by each Web server existing on the Internet
The file has a function of collecting HTML files.
Collection can be performed by an eb robot. The Web robot randomly collects and accumulates HTML files managed by many WWW servers so as to be searched by an HTML file search system while sequentially moving the WWW servers. W owned by WWW server
Many web pages are updated daily, so Web
The robot periodically visits the WWW server to detect an updated Web page.

Thereafter, the HTML about the newly added information and the updated information collected by the Web robot will be described.
When the URL of the file is received, the received system accesses the corresponding HTML file based on the URL and obtains the document information. By performing the same operation on the obtained document as in the case of the claim data 107, necessary information can be selected. In addition, to supplement the contents of the sample data in the case of this embodiment,
W which posted the topic that the user wanted as positive data
Give dozens of HTML files of web pages. Similarly, as the negative example data, dozens of HTML files not describing the topic desired by the user are given. However, in this case, as the negative example data, it is desirable to select an HTML file having a close topic without being completely unrelated to the contents of the positive example data.

For example, if the user wants to know an HTML file having the topic of singer A, an HTML file containing the topic of singer A
The L file is used as positive example data, but as negative example data, information such as singer B or singer C is selected. An analogy rule is generated based on the above sample data. When a topic other than singer A, which is a desired topic, is received, a trash box or the like may be disposed as a distribution destination so as to discard the topic.

[0050]

The first effect is that a keyword for determining a distribution destination is not given by a person in advance, but a complaint or Q & A mail which has been distributed in the past is prepared for each distribution destination and a distribution destination is determined. By using the sample data to be used, a word characterizing the distribution destination is automatically selected therefrom to form an analogy rule for selecting the distribution destination. The contents of the complaint are automatically classified according to the analogy rule with respect to the newly received mail. If the address of the person in charge is assigned to each category, for example, a vacuum cleaner complaint is directly sent to the person in charge of the vacuum cleaner, which contributes to labor saving in a department that receives a great deal of information.

The second effect is that when the similarity is lower than the threshold value and the distribution destination cannot be determined, or when the wrong distribution destination is selected, the claim data is input as sample data for the correct distribution destination and re-learning is performed. And review the previous analogy rules. This re-learning can also be automated.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of the present invention.

FIG. 2 is a sample example of a claim for explaining pre-processing of an embodiment of the present invention.

FIG. 3 is an example of a claim to be distributed for explaining a classification process according to the embodiment of the present invention.

FIG. 4 is an example of decomposing sample data into words in preprocessing according to the embodiment of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 101 Analogue rule generation part 102 Input part 103 Natural language analysis part 104 Classification part 105 Sample data for analogy rule generation 106 Analogy rule file 107 Claim data 108 Text file 109 Composition word data file 110 Classification result

Claims (10)

[Claims] 1. A pre-process for generating an analogy rule for classifying documents, and a classification for classifying and distributing complaint data including claims and inquiries by a person in charge using the analogy rule generated in the pre-process. The pre-processing is performed by inputting an analogy rule generation sample data file that stores claim data for all distribution destinations whose distribution destinations are already known, and inputting the analogy rule generation sample data file. An analogy rule generation unit that generates an analogy rule consisting of a word or a word string for a distribution destination, and an analogy rule file that stores the analogy rule generated by the analogy rule generation unit,
The classification process includes inputting the claim data and converting it into a text format, a text file storing the claim data converted into the text format, and inputting the claim data converted into the text file and performing morphological analysis. A natural language analysis unit for outputting as constituent word data composed of decomposed words, a constituent word data file storing the constituent word data, and the analogy rule file for constituent word data input from the constituent word data file. A classification unit that calculates a similarity by applying the analogy rule to each distribution destination, and distributes the claim data to the distribution destination having the highest similarity. 2. When generating an analogy rule in the preprocessing, an analogy for deciding a distribution destination by inputting claim data which has already been identified for all distribution destinations and pre-sorted for each distribution destination. 2. The automatic document information evaluation apparatus according to claim 1, wherein rules are generated for all distribution destinations. 3. The automatic document information evaluation apparatus according to claim 1, wherein the claim data includes any of electronic mail, voice information by telephone, voice mail, and fax information. 4. When generating analogy rules in the pre-processing, when distribution destinations have already been identified for all distribution destinations and claim data preliminarily classified for each distribution destination is input, claim data distributed to a certain distribution destination is obtained. It is classified into two with the claim data distributed to other than the destination, and the ratio of the appearance frequency of each word or word string included in the two classified claim data is calculated for all of the claim data of the destination. 3. The document according to claim 1, wherein a word or a word string is obtained, an analogy rule for the destination is generated by comparing the word or word string with a predetermined threshold, and the rule is performed for all the destinations. Automatic information evaluation device. 5. In the classification processing, the complaint data is converted into text format data, and the constituent word data obtained by decomposing the text data into a word string by morphological analysis is compared with the analogy rule of each distribution destination. Is executed for all the distribution destinations, the similarity is calculated, and when the maximum value calculated is equal to or greater than a predetermined threshold, the claim data is distributed to the distribution destination that has calculated the maximum value, and the threshold is calculated. 2. The automatic document information evaluation apparatus according to claim 1, wherein the distribution is not performed when the value is less than the value. 6. The automatic document information evaluation apparatus according to claim 1, wherein the analogy rule generation sample data file includes claim data or constituent word data corresponding to the claim data. 7. In claim data distribution, claim data not delivered below the threshold value or claim data delivered to an unauthorized delivery destination are sampled data of a delivery destination that should be delivered. Additional registration in the analogy rule generation sample data file,
2. The automatic document information evaluation apparatus according to claim 1, wherein the learning is performed by activating the analogy rule generation unit and revising the analogy rule. 8. When the claim data is received, the input unit is automatically started, and when the claim data which is not delivered below the threshold value or the claim data delivered to an illegal delivery destination is generated, the original delivery is performed. 2. The analogy rule generation sample data file is automatically added and registered as sample data of a distribution destination to be performed, and the analogy rule generation unit is automatically started to automatically review analogy rules. Or the document information automatic evaluation device according to 7. 9. E-mail, telephone, voice mail, fax
In the system that automatically distributes the claim data received by the user to the person in charge, the claim data for which the distribution destination is already known is input, the analogy rule composed of words or word strings is obtained for all distribution destinations, and newly received. After converting the claim data into a text format, a morphological analysis is performed to create constituent word data composed of decomposed words, and a similarity rule for each of the distribution destinations is applied to the constituent word data to calculate a similarity. A document information automatic evaluation system for distributing to a distribution destination having a maximum similarity. 10. An HT on which topics for which acquisition is requested are posted.
Generating analogy rules by using ML files and HTML files that do not include topics that are requested to be obtained as sample data, and generating HTMs such as Web robots on the Internet
10. The automatic document information evaluation system according to claim 9, wherein the latest HTML file is obtained from the L file search system, and an HTML file having a topic desired to be obtained is searched based on the analogy rules.