JPH0782503B2 - Article integrated processing device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial application] The present invention relates to a document database device,
The contents of multiple articles generated from the same news source are integrated to provide complete information as one article. In particular, it is used in a system that provides readers with integrated information of a plurality of newspaper articles written based on the same news source as one article.

〔Overview〕

According to the present invention, in an article integration processing device that integrates and organizes a plurality of article contents described for the same target and provides them as one integrated article, the extracted headline has the largest amount of information. From the beginning, the headings with the same contents are rejected, and the integrated sentence adopts the article with the most information as the basic article. For each sentence of the basic article, about the sentences of other non-adopted articles and its information amount By creating a unified sentence by replacing the sentence with the larger amount of information by comparison, and extracting sentences with contents not included in the created unified sentence from the non-adopted articles and adding them It is intended to provide a comprehensive source of information about the content of the source, without having to read multiple articles.

[Conventional technology]

In newspapers or magazines, each reporter writes articles from one angle for one news source. Therefore, the articles written by the newspaper companies based on the same news source have different contents, and it is said that you should read multiple newspapers to grasp the situation accurately.

Therefore, businessmen are required to order and read a plurality of newspapers. In addition, there is a business that cuts out a requested subject for multiple newspapers and provides it to the readers who have contracted it.

[Problems to be Solved by the Invention]

However, there is a problem that it is easy to waste time and money to order multiple newspapers and examine the information described therein, and the method of contracting to receive articles is a heavy burden for individuals, and There is a problem that it takes manpower to cut out.

Also, each newspaper company provides a database of the article contents of newspapers it provides. However, since each newspaper company only provides its own articles for this database service, if you want to consider articles from multiple companies, you need to make a contract for providing a database for each company, Considering multiple articles has been a problem of high cost and labor.

The present invention is to solve the above-mentioned problems, and to provide an article integration system capable of automatically creating and providing one integrated article without losing information from a plurality of articles. To aim.

[Means for Solving the Problems]

The present invention provides an article integration processing device for storing a plurality of article contents describing the same target in a file and creating one integrated article from the original article contents stored in the file, A text extraction unit that extracts the text of articles to be integrated from a file, and an article text that has a large number of characters from the extracted article texts and that has a large number of important words that are words and phrases related to the subject of the article text. The basic article determination means adopted as the basic article sentence of the integrated sentence and the number of characters of each sentence included in this adopted basic article sentence are compared with the text of other articles other than the basic article describing the same content to determine the number of characters. If there are many sentences, replace the sentence of the basic article to create an integrated text, and the created integrated text is divided into reading points. For each, it is judged whether all the same words as the words already appearing in the integrated text located before the sentence to be processed are included, and if all are included, the sentence to be processed is The basic article determining means includes means for performing weighting from an article having the largest number of characters, and important words included in the article, including integrated text processing means having a sentence shaping means for deleting and shaping the integrated text. And a means for adding weights by the number of characters and the number of important words and adopting the article with the largest number of weights as a basic article. .

In addition, the sentence replacement means of the integrated text processing means cuts out the numbers and important words in each sentence in the basic article, compares them with the sentences of other articles containing the cut-out numbers or important words, and selects the longer one. It is preferable to provide means for adopting a sentence as an integrated text.

In addition, a means for extracting the headings of articles to be integrated from the above file, a means for sequentially selecting the heading with the longest number of characters from the extracted headings, and a character included in the selected heading It is preferable to include integrated headline processing means including headline identity determination means for rejecting headlines having the same content by comparing the characters included in the adopted headline.

Further, the headline identity determining means of the integrated headline processing means is a cutting-out means for cutting out a number or an independent word from a character in each heading, and a cutting out means or an independent word if there is a number or an independent word. When a word is in an already adopted heading, the means of disabling the heading containing the number or its independent word is compared to the characters in the already adopted heading, and more than a certain number of characters match. It is preferable to include a heading that has not been adopted and a heading in which the number of mismatched characters is a predetermined number or less.

Further, in the article integration processing device of the present invention, a means for cutting out the important words and numbers from the integrated article body, and a sentence in which the cut out important words and numbers are not present at all, is extracted from the original article sentence, And a means for adding to the integrated headline and the integrated sentence.

[Action]

Multiple articles are stored in the file as integrated articles.

For the headline of the integrated article, first, the entire headline is extracted, and the headline having the largest amount of information is adopted as the heading of the integrated article. At that time, headings with the same content are rejected and only headings with different content are adopted.

Follow the steps below to create the text of an integrated article. First, we select the sentence that is considered to contain the most words and the most important words from multiple articles as the basic article of the integrated sentence. And, for each sentence of this adopted basic article,
Compared with the sentences of other articles that describe the same content, the sentence that is considered to have more information is adopted, the sentence as a whole is shaped, and the text of the integrated sentence is created.

Furthermore, sentences that are not adopted in the above-mentioned integrated text and have completely different contents are extracted and added as special notes to the headline and the integrated text to create an integrated article as a whole.

〔Example〕

 Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of an article integration processing device according to an embodiment of the present invention. In this embodiment, an example of a system that creates one integrated article from a plurality of newspaper articles will be described.

The article integration processing apparatus of this embodiment is roughly divided into four parts. That is, the file device 1 in which the original articles to be integrated are stored, the integrated headline processing means 2 for processing the original article headlines stored in the file device 1 to create an integrated headline, An integrated text processing means 3 for creating an integrated text integrated from the original article in the file device 1 and the created integrated heading and important information not included in the integrated text are extracted into the text. It is composed of special note processing means 4 to be added, and creates an integrated article 5 integrated from original newspaper articles.

Next, the operation of each means will be described.

The file device 1 is provided with an original article file 11, and the original article file 11 stores original articles provided by a plurality of newspaper companies about one target to be integrated.

That is, articles written based on one news source in each newspaper are extracted and stored in the original article file 11. This article is stored by normal character recognition processing, the article content is classified according to the content of the recognized article, and articles related to the same news are designated for the integration processing.

Next, the integrated headline creating process in the integrated headline processing means 2 will be described.

The integrated headline processing means 2 includes a headline extraction means 21 for extracting only headlines from original articles of newspaper articles across a plurality of companies to be integrated in the original article file 11, and the extracted headings have a large number of characters. The heading selecting means 22 that selects from the order and the heading selected by this heading selecting means 22 determine whether or not there is the same content as the heading already adopted, and it is the same as the heading already adopted. When the heading has the same content, the heading identity judgment means 23 that rejects the integrated heading as rejected is provided, and finally the heading of the integrated article is determined by removing the heading of the same content from the extracted heading. To do.

Next, the operation of the headline identity determination means 23 will be described with reference to the operation flowchart of FIG. This headline identity determination means 23 rejects the headings that are selected as the headings selected in descending order of the number of characters if they have the same content as the headings already adopted for each heading as unnecessary. This is a process that is rejected as an integrated heading.

That is, as shown in FIG. 2, it is determined whether or not the heading selected as the integrated heading has a number (step S21). If the number is included, the number is cut out (step S22). If the number is a decimal point, it is cut out together, if it is an Arabic numeral, the Chinese numeral is also cut out, if it is a Chinese numeral, the Arabic numeral is also cut out. Cut out. Then, if the same number is cut out from the remaining headline that has been adopted, it is regarded as the same content, and this heading is rejected as an unadopted integrated heading (step S2).
3, 26). Further, the remaining headlines that have been adopted are compared with the characters, and if seven or more characters match or if the mismatching characters are two characters or less, they are rejected as unaccepted as having the same content (steps S24 and S26).

If the selected headline does not have a number, a word composed of kana or kanji enclosed in hiragana is cut out (step S28). In this word cutout, two or more characters are cut out as a word, words at the beginning and end of a sentence are cut out at a hiragana on one side, and spaces in the heading and special codes are treated as hiragana. Then, this cut-out word is compared with the words of the remaining headings that have already been adopted, and if there are the same cut-out words in the remaining headings, they are rejected as rejected because they have the same content (step
S23, 26). Further, the remaining headings that have been adopted are compared with the characters, and if seven or more characters match or if the mismatching characters are two characters or less, the same content is rejected and rejected (steps S24, S26). ).

In this way, while adopting an integrated heading from a heading having a large number of characters, the headings having the same contents are rejected to determine the integrated heading (step S25).

A specific example of this integrated headline creation processing will be described.

The following headings extract the headings of the original text articles of each newspaper company regarding the high-speed document retrieval system developed by Hitachi, Ltd., which was published in the issue of July 26, 1989.

Radio newspaper one year of locating in the newspaper sandwiched 1.5 seconds document retrieval system developed by Hitachi ultra-high-speed type Asahi Shimbun one year's worth of articles Hitachi ultra-high-speed retrieval of system development two years after the prospect commercialization Nikkei Sangyo you want to find in 1.5 seconds from newspaper Quickly search documents with newspaper WS Hitachi low-cost system prototype Nihon Keizai Shimbun Hitachi 1.5 seconds required document search WS prototype system Denki Shimbun Search all words Hitachi ultra-high-speed document search system Nippon Kogyo Shimbun Articles in one year in 1.5 seconds Hitachi's high-speed search system Yomiuri Shimbun Newspaper articles searched in 1.5 seconds for one year Hitachi new system Nikkan Kogyo Shimbun Newspaper articles for one year searched in only 1.5 seconds This headline was first searched for in the Nikkan Kogyo Shimbun Search for newspaper articles in just 1.5 seconds ”is selected and adopted as the first integrated heading. Next, heading with the largest number of characters, "Hitachi, Ltd. Ultra-high speed search system development"
Even if you select and cut out a word, there is no same word in the previous heading, so it is adopted as an integrated heading. The next headline with a large number of letters is "Search for a newspaper article in 1.5 seconds in a year", but the heading has the numbers "1 year" and "1.5" cut out, and this number is the first to be adopted first. Since the same word is included in the heading, it is rejected as having the same content. In this way, all headings after the ones containing "1 year" and "1.5" are rejected.

Also, the headline of "Hitachi Ultra High-speed Document Search System" in Denki Shimbun is rejected because it has 7 or more characters that are the same as the headline adopted second.

In this way, from the above-mentioned headlines, the headline identity determination means 23 adopts and determines eight integrated headlines as the integrated headings as shown in the integrated headline shown in a separate sheet.

Next, the operation of the integrated text processing means 3 will be described with reference to FIGS.

As shown in FIG. 1, the integrated body text processing unit 3 extracts the body text of the original article file 11 from which the body text is removed, and the body text extraction unit 31 extracts the body text of each article. , A basic article determination means 32 for performing weighting based on the number of characters and the number of important words extracted from each article text, and determining the article text with the largest weighting number as the basic article text, and this basic For each sentence of the basic article decided by the article deciding means 32, compared with the sentence of the other article text describing the same content,
The sentence replacement unit 33 that selects a sentence with a larger amount of information and has a richer content and adopts it as an integrated text, and the sentence shaping unit 34 that shapes the entire sentence to make it smooth as a sentence, The integrated text of the integrated article 5 is created.

First, the operation of the basic article determining means 32 will be described based on the flowchart of FIG.

When integrating the article texts, it is first necessary to determine which article is used as the basis for creating the integrated text, and the basic article determining means 32 determines this. The basic idea of this basic article decision judgment is to extract an article with a large amount of information and a high content as a basic article.

First, only the body of the article to be integrated is extracted from the original article file 11 by the body extracting means 31 (step S31). The number of characters in the extracted article body is counted, and the article number is reversed from the article body having a large number of characters, that is, the weighted number. (Step S32). Next, the number of important words in the article body is calculated (step S33). This important word is
It refers to a word composed of the particles "ha", "ni", "toha" and the kanji / kana character string before the reading point "," from the beginning of the sentence. Here, the word before this particle is taken as an important word because this word constitutes the subject of the sentence,
The words that use "ha", "ha", and "toha" are considered to correspond most to the concept that the sentence is trying to describe, so such words are extracted and their frequency of expression is counted. By doing so, it is possible to know the richness of the content.

Therefore, according to the number of important words, the articles with the most important words are weighted and the number of articles is reversed (step
S34). Then, the weighted number of characters and the weighted number of important words are added to adopt the article body with the largest weighting as the basic article body (step S35).

A specific example will be described.

The following table shows the weight of the number of characters and the important words in the article text of each newspaper company regarding the high-speed document search system developed by Hitachi, Ltd., which created the above integrated headings.

Therefore, in this example, the text of the article in the Denpa newspaper is adopted as the basic article.

Next, the processing operation by the sentence replacement means 33 is shown in FIG. The processing by the sentence replacement unit 33 is processing for adopting a sentence having a larger amount of information as a unified text for sentences expressing the same content, and a sentence having a larger amount of information for the same content than the basic article is different. If there is an article, it is a process for replacing the sentence of the basic article.

First, the basic article is cut out sentence by sentence and each sentence is read (steps S41 and S42). Next, it is judged whether or not the sentence includes a number (step S43), and the number is cut out (step S44). Furthermore, it is judged whether or not there is an important word in the sentence (step S45), and if there is an important word, the important word is cut out (step S46), and the complemented sentence including all of the cut out important word and number Is searched from the text of other articles (step S47). The lengths of the supplemental sentence and the basic article body are compared (step S48). If the supplementary sentence is longer, it is replaced with the basic article body and adopted as the integrated body sentence (step S49). At this time, when the complementary sentence has already been adopted as the integrated text, it is compared with the next candidate sentence and replaced with a complementary sentence having more characters than the basic article text.

If there is no important word, search for a supplemental sentence containing all the cut out numbers from other article texts (step 50), compare this supplemental text with the basic article text and the number of characters, and integrate the longer sentence. Is adopted (steps S48, S49).

If there is no number in the basic article text but an important word,
An important word is cut out, and a complementary sentence including the word is searched for from other article texts (steps S51, 52, 53). And
If there is a complementary sentence longer than the basic article sentence by steps S48 and S49, the complementary sentence is adopted as the integrated text. If there is no important word, the basic article text is adopted as the integrated text (step S54).

In this way, all the sentences in the basic article text are examined for the same content sentence by sentence, and the sentence considered to have the largest amount of information is adopted as the integrated text.

A specific example will be described.

The article on the high-speed document retrieval system developed by Hitachi, Ltd. was adopted by Denpa Shimbun as a basic article.

The first sentence of the text of the basic article is "Hitachi, Ltd. announced that it has developed an ultra-high-speed document retrieval system that puts a large amount of document information into practical use in free-word retrieval. ].

From this sentence, "25" is cut out as a number and "Hitachi" is cut out as an important word. The same supplementary sentence as these two letters was extracted from the articles of other companies, and the number of words was compared. "Hitachi Ltd. (Katsushige Mita, President) 25th, electronic digitization created by warp etc. of Nikkan Kogyo Shimbun It announced that it has developed an ultra-high-speed document retrieval system (photograph) that can retrieve documents using any of the words in the text. It turns out that the number of characters is larger. For this reason,
The supplementary text of this Nikkan Kogyo Shimbun is adopted as the integrated text and replaced with the basic text.

In this way, the sentence replacement means 33 compares each sentence of the basic article text with the complementary sentences of other articles, and takes in a sentence having a large amount of information as an integrated text.

In the above article example, seven sentences were replaced with complementary sentences.

Next, the processing operation of the sentence shaping means 34 will be described with reference to the flowchart of FIG.

In the sentence replacement means 33, the basic article text is replaced with the complementary text of another article, so that sentences with different contexts are mixed in the integrated text, and sentences with the same content exist in the integrated text. For this reason, it is necessary to perform text formatting to smooth the entire text of the integrated text.

In this sentence shaping, first, a word string is cut out in reading point units from the first line of the integrated text (step S61). Next, from the cut out word string, words of hiragana, special symbols, kanji surrounded by spaces, katakana, and numerical words are cut out (step S62). However, if it is one character, it is ignored. And
From the next line, when all of the words cut out in step S42 are included in reading point units, the word string is deleted in reading point units and the sentences are packed (step S63).

A concrete example of this will be described in the article on the high-speed document retrieval system developed by Hitachi, Ltd.

According to Denpa Shimbun, which became the basic article text, "In the case of a general technical document consisting of about 10,000 characters, 25,000 can be searched in about 5 seconds. 』There is a sentence. Even if this sentence is compared with the complementary sentence by the sentence replacement means 33, the sentence containing this number has been adopted first, and there is no sentence to be the next candidate. Remain in. However, in the previous integrated text, the sentence replaced from the supplemental sentence of Nikkan Kogyo Shimbun says, "The system has a maximum read speed of 20 MB per second RA.
M disk, aggregate magnetic disk drive of 10 Mbytes per second, 1. 6M bit search processor,
With a string search processor of 20 Mbytes per second, the system achieves a system search speed of equivalently 100 Mbytes per second, and in the case of a general technical document of about 10,000 characters, 25,000 items can be obtained in 5 seconds. You can search. It can be used for patent information / literature information service, article information management for magazines, newspapers, etc. , And there is a word string containing all of the same words mentioned above, and the sentences have the same content, so the sentences in the basic article body are deleted.

Through such processing, the integrated text described in the attached sheet is created.

Next, the remarks of special notes will be described.

Other article sentences not adopted in the above-mentioned integrated text may include information different from the basic article sentence. In order to add information lacking in the integrated text to the integrated article sentence, it is necessary to perform processing to extract a sentence whose content is completely different from the integrated text from articles other than the basic article. This processing is performed by the special item processing means 4.

FIG. 6 shows the processing by the special note processing means 4.

First, important words and numbers are cut out from the integrated text (step S71). The extraction of the important words and the numerical words is performed on the same basis as the extraction of the important words and the numerical words performed by the integrated text processing means 3.

Then, a sentence in which the cut-out important words and numbers do not exist at all is extracted from the original article sentence and added to the integrated headline and the integrated text as a special note for each newspaper company (step S72).

In the article about the high-speed document retrieval system developed by Hitachi, Ltd., as a special note, for example, the Asahi Shimbun said, "In Hitachi, data is first checked by letters, then words, documents, and then three-step sifting to check the data. It adopted the method to find the sentence of, and shortened the time. For example, in the case of a sentence "information is being commercialized," this sentence is first searched for in a list that is decomposed into individual letters "information,""report,""no," and then "information, commercialization ..." Check the "condensed sentence" without the above and narrow down the candidates. "Hitachi's newly-developed method requires less man-hours for storage and can be easily searched. "The company says that you can search at the same speed as or faster than the conventional method. ”The four sentences of“ Hitachi's words that were thought to be unnecessary as keywords when inputting may become important after 5 years or 10 years. ”Were extracted as special notes.

As the important words in the above-mentioned integrated text, "Chosita", "25", "Hitachi", "1 year", "72 million", "1.5"
"500""Ultra-high-speedsearch""1300""Synonyms""Te3dan""1000""1ststage""System""2ndstage""3rd""25,000""Ba" primary "5,000"",secondary""100,000""about 10 times""aggregate magnetic disk device""thistime""20""high speed multiple character string collation method""second10M""one."
"Six", "One hundred M", "10,000", "Five seconds", "Same system"
"About 5 seconds" and "27" are extracted.

For this reason, "Hitachi", which is extracted as an important word in the above-mentioned special mention sentence of the Asahi Shimbun, is not extracted as an important word in the integrated text, and "5 years" and "10 years" are also extracted. Since there is no such information, it is determined that the information related to the above sentence is not in the integrated text.

In this way, an integrated article consisting of the integrated headline, the integrated text, and the special notes for newspapers is created and provided to the reader, and the reader can read an information-free article by reading one integrated article. You can get information from multiple newspapers in your own room.

〔The invention's effect〕

As described above, since the present invention can automatically provide articles to a reader by integrating articles of a plurality of newspaper publishers as an integrated article with no information leakage, the reader can comprehensively review a review by multiple writers. You can grab and know the exact information.

Further, since the reader himself can know the information related to one news source without hitting all newspapers, the labor and cost for that can be reduced.

Attachment (Example of created integrated article) Integrated headline Asahi Articles that you want to find in 1.5 seconds from a year's newspaper Asahi Hitachi Ltd. System development for ultra-high speed search Asahi "No need to register headline" Nissan Hitachi Low cost system prototype Nikkei WS adopted system prototype in the quick document search Yomiuri Hitachi new system Asahi two years later prospect market integration-text Hitachi, Nissan WS (President Mr. Katsushige Mita) 25th, electronic written in, for example, a word processor It announced that it has developed an ultra-high-speed document retrieval system (photo) that can retrieve documents using any of the words in the text. You can search about 72 million characters in a yearly newspaper article with the same system in about 1.5 seconds.

The ultra-high-speed search by the full-text search method this time was realized by the combination of the following technologies developed by the Central Research Institute, which resulted in an integrated speedup of about 500 times as a system.

When there are discrepancies in expressions among synonyms, different notation words, etc., we automatically searched based on 1,300 different notation conversion rules, and cut off the particle part and repeated words of sentences. By combining the hierarchical pre-search method, which searches by text, etc., the overall search efficiency is improved.

Character components tables that have been created and compressed automatically from the text of the document in advance, and annexes such as particles and conjunctions are deleted from the text, and condensed text that eliminates repeated words The original method, such as the hierarchical search method, which performs the search to increase the search speed is adopted to overcome the shortcomings of the full-text search method and be put to practical use.

The system developed this time automatically creates up to 1000 kinds of synonyms and different notation words when a certain word is input from the terminal and performs the first stage search.

Then the second stage selection begins with a compressed sentence that removes particles and conjunctions from the sentence. This narrowing down made the final selection easier.

The third step is a direct search by text search.

If the original data is 25,000, it can be narrowed down to 5,000 in the primary search and 500 in the secondary search. After this, make a final search.

The read speed of the document data is doubled by arranging the collective magnetic disk device and N compact devices in parallel and reading the document data in parallel and independently. This time, 12 units were lined up to achieve about 10 times faster speed.

High-speed multi-character collation method, which searches and inquires at the same time for a maximum of 1,000 synonyms and different notations in one scan of document data, and uses a dedicated processor to achieve a collation speed of 20 Mbytes per second. Realization.

The system has a maximum read speed of 20 Mbytes RAM per second
Disk, aggregate type magnetic disk device of 10 Mbytes per second,
With a bit search processor of 1/6 Mbits per second and a character string search processor of 20 Mbytes per second, the configuration, equivalently achieves a system search speed of 100 Mbytes per second, of the general technical documents of about 10,000 characters In that case, 25,000 items can be searched in 5 seconds. It can be used for patent information / literature information service, article information management for magazines, newspapers, etc.

In addition, this information was announced at the 27th Information Processing Society of Japan.

Additional items "Nikkan Kogyo" Unlike the conventional search method, which required indexing each document, a new method that allows you to freely specify and search any word by directly referring to the contents of the document method.

"Nissan" Plan to commercialize within two years "Nissan" Data is stored in a fixed disk unit (HDD) of 5.25-inch size. Multiple units are connected in parallel.

"Asahi" Hitachi adopted a method to search for the target sentence by first checking the data through a three-step sieving, first with letters, then with words, then with sentences, to shorten the time. For example, in the case of a sentence "information is being commercialized," this sentence is first searched for in a list that is decomposed into individual letters "information,""report,""no," and then "information, commercialization ..." Check the "condensed sentence" without the above and narrow down the candidates.

"Asahi"Hitachi's newly developed method requires less man-hours to store, and can be easily searched.

"Asahi" The company says that you can search at the same speed as or faster than the conventional method.

"Asahi" At Hitachi, "Words that were thought to be unnecessary as keywords when inputting may become important words in five or ten years.

"Electrical" Unlike the conventional search method that required indexing each document, a new method that allows you to freely specify and search for any word by directly referring to the content of the document is adopted. There is.

"Electricity" The company wants to put a product into commercialization within the next two years.

"Nikkei" Commercialize and release within two years.

"Nikkei" Data is stored in a fixed disk unit (HDD).

"Yomiuri" In the previous methods, the index attached to each article (I had to search based on the index word,
The new method is a mechanism that allows you to quickly retrieve the desired article part by combining multiple arbitrary keywords specified by the user without using an index.

"Yomiuri" Hitachi is aiming for commercialization in two years.

"Nippon Kogyo" Aiming for commercialization two years later, the price has not been decided.

[Brief description of drawings]

FIG. 1 is an overall configuration diagram of an embodiment of the present invention. FIG. 2 is a flowchart for explaining the headline identity determination means of the integrated heading processing means according to the embodiment. FIG. 3 is a flowchart for explaining the basic article determining means of the embodiment integrated text processing means. FIG. 4 is a flowchart for explaining the sentence replacement means of the embodiment integrated text processing means. FIG. 5 is a flowchart for explaining the sentence shaping means of the embodiment integrated text processing means. FIG. 6 is a flow chart for explaining the special note processing means of the embodiment. 1 ... File device, 2 ... Integrated headline processing means, 3 ... Integrated text processing means, 4 ... Special note processing means.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-1-112331 (JP, A) Communication Society of Japan Material VOL. CMN-89, No. 18-23 (1989-7-12) P. 51-60

Claims (5)

[Claims] 1. An article integration processing apparatus for storing a plurality of article contents describing the same object in a file and creating one integrated article from the original article contents stored in this file, A body text extraction means for extracting the body texts of articles to be integrated from the above files, and an article body body having a large number of characters from the extracted plurality of article body bodies and having a large number of important words that are words and phrases related to the subject of the article body body. The number of characters compared to the basic article determination means that adopts as the basic article sentence of the integrated sentence, and the number of characters of each sentence included in the adopted basic article sentence compared with the text of articles other than the basic article that describes the same content. If there is a sentence with a lot of sentences, the sentence replacement means that replaces the sentence of the basic article to create the integrated text, and the created integrated text is divided into reading points, and the divided text For each of the above, it is judged whether or not all the same words as the words that have already appeared in the integrated text located before the sentence to be processed are included. If all are included, this sentence to be processed is deleted. The basic article determining means includes means for weighting the article with the largest number of characters and important words included in the article. Articles characterized by having means for extracting and weighting the articles with the largest number of important words, and means for adding the number of weights by the number of characters and the number of important words and adopting the article with the largest number of weights as a basic article Integrated processing device. 2. The sentence replacement means cuts out the numbers and important words in each sentence in the basic article, compares them with the sentences of other articles containing the cut out numbers or important words, and integrates the longer sentence. The article integration processing apparatus according to claim 1, further comprising means for adopting the article as a text. 3. A means for extracting headlines of articles to be integrated from the file, a means for sequentially selecting headlines having the longest number of characters from the extracted headlines, and a means included in the selected headlines. 3. The integrated headline processing means comprising: headline identity determination means for rejecting headlines having the same content by comparing the characters with the characters included in the headlines already adopted.
The article integrated processing device described. 4. The headline identity determination means cuts out a number or an independent word from a character in each headline, and, if there is a number or an independent word, the same number or the same independent word has already been adopted. Means to reject a heading containing the number or its independent word when it is in a headline that has been selected, and a headline that does not match the number of characters in the headline that has already been adopted 4. The article integration processing device according to claim 3, further comprising means for rejecting a headline in which the number of characters is less than or equal to a predetermined number. 5. A means for cutting out important words and numbers from the integrated article body, and a sentence in which the extracted important words and numbers do not exist at all are extracted from the original article sentence to obtain the integrated heading and integrated sentence. 5. The article integrated processing device according to claim 1, further comprising: