JP2000242652A - Information stream retrieval method and device and storage medium recorded with information stream retrieval program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To retrieve the transitions of objective topics from the transitions of the topics of a document extracted from an information transmission medium and to sequence and provide them in an order suited to a retrieval request. SOLUTION: In a document storage part 102, document information inputted in a document input part 101 is stored. In an information stream extraction part 103, classification by time information and classification by topics are executed to a document set inputted in the document input part 101, the documents classified into individual topics are arranged by using the time information and an information stream is generated. In a retrieval request input part 104 the document in which a word desired to be retrieved or information desired to be retrieved is written is specified from the extracted information stream and the retrieval request is originated. In an information stream retrieval part 105, the documents or classification categories satisfying a retrieval condition are extracted from the information stream extracted in the information stream extraction part 103, the similarity with the retrieval condition of them is calculated and they are displayed at a display part 106 sequentially as a retrieved result starting from the information stream of high similarity.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an information flow search method and apparatus for extracting topics and temporal transitions from a large amount of information transmitted from an information transmission medium and searching the information.

[0002]

2. Description of the Related Art In recent years, it has become possible to exchange a large amount of electronic documents through a computer network such as the Internet. For this reason, services that allow individuals to search for information they need are implemented on networks. However, the amount of information acquired by the user becomes large, and it becomes difficult to extract characteristics of individual information. Therefore, a technique for classifying and organizing the acquired information is required.

Conventionally, research has been conducted on a method of automatically classifying document information. A typical method is to classify information into categories that are considered appropriate for new information, with known breaks (called categories) for classifying like a library (" Automatic text classification "Yamamoto, Masuyama (Toyohashi University of Technology) Naito (NTT), 1995), or a method in which the category to be classified is unknown and similar documents are collected from a set of documents to create and assign a classification category (" Automatic division by a competitive learning neural network "Kikuchi, Matsuoka et al. (Utsunomiya Univ.), 1995). With these technologies,
Classifying and organizing a large number of documents.

Further, as described in the “information flow presentation device” (Japanese Patent Laid-Open No. 10-154150) proposed by the present applicant, a topic such as a newspaper article having time information is targeted. There is also a technique for extracting the transitions of. This technology automatically collects documents with similar topics from a set of documents, creates categories, divides them at certain time intervals, and presents them. It is to extract.

FIGS. 5 and 6 show an information flow presentation device (Japanese Patent Laid-Open No.
No. 0-154150) to extract topics from newspaper articles. FIG. 5 shows an information flow relating to “Typhoon No. 3”, and FIG. 6 shows an information flow relating to “Elementary school murder”.

In FIG. 5, the date “1998/06/08”
"Typhoon No. 3" was sent as a topic, and the date "199
"8/06/09" indicates that "power expansion" of Typhoon No. 3 is extracted as a topic. In addition, "Typhoon No. 3" indicates that two articles are transmitted, and "Typhoon No. 3""KinkiPass" indicates that six articles are transmitted. By the way, the empty category on the date "1998/06/07"
"The information tide is starting on 1998/06/08"
It represents that. Further, FIG. 6 shows that various topics are present in the information flow of “elementary school murder”. For example, the date “1998/06
/ 28 ”, a“ murder case ”occurred, and the date“ 1998/0
In June 29, the topic of the killed elementary school students and the doctor
It shows that the topic is divided into the topic of diagnosis and the topic of the junior high school of the culprit.

That is, the information flow presentation device collects similar articles from a set of documents transmitted at each date or time, assigns them to a category, determines a category name, and determines a similar topic existing at each time. It can be said that it is a device that connects categories and presents the results.

[0008]

According to the invention described in the information flow presenting apparatus (JP-A-10-154150) previously proposed by the present applicant, if the desired information is known in advance, the information is used as the information. By inputting the information into the tidal current presentation device, the transition of topics according to the information could be extracted.

Another object of the present invention is to provide an information flow presentation method and apparatus for searching for a necessary topic transition from a large number of topic transitions extracted from an information transmission medium, and presenting the transition in a sequence. That is.

[0010]

[Means for Solving the Problems] First, a topic transition from an input article set (this is called an information flow)
Is extracted. For this purpose, information flow is extracted. The document to be classified is a document having time information, and uses the time at which the document was created and the time expressed in the document. The collection of articles uses information transmitted from a medium that transmits text information, such as a newspaper company, a radio station, or a television station, separately or in combination. The temporal change of the topic is obtained by connecting the common topic categories in chronological order using the result of classifying the article set by time and the result of classifying by topic.

Before describing the information flow search technique, a feature vector is defined for each document to be searched. The feature vector is a real-valued vector, and each element has one-to-one correspondence with each word existing in the document. Feature vector of document i

[0012]

[Outside 1] Is

[0013]

(Equation 1) Can be expressed by n is the number of words used to represent the characteristics of the document. The value of each element is determined using the appearance frequency of the word, the distribution ratio in the entire document set, the length of the number of characters, etc. ("Automatic Text Processing" Gera
rd Salton, ADDISON-WESLEY pub. 1989) is common. The similarity between documents can be defined using this feature vector. For example, a function such as cos θ (where θ is the angle between the feature vectors) calculated from the inner product between the feature vectors is used.

Next, a method of searching for an information flow will be described. The search request uses a word or a document in which information to be searched is written. Each search request is specified by one, a plurality, or some combination. If the search request is a word, all documents containing that word are selected for all documents. At this time, the value of the element of the feature vector of the document is used for the similarity between the word of the search request and the document. If the search request is a document, a feature vector of the document is generated, the similarity to the feature vector of the search target document is calculated, and all similar documents are selected.

A document set satisfying a search request is defined as SDoc =
{Doc ₁ ,..., Doc _m }, each similarity is expressed as
oc v = (doc v ₁ , ..., doc v _m ).
m is the number of documents that satisfy the search request. Next, the similarity of each information flow with respect to the search request is calculated. For example, when the similarity Rel _k of the information flow _k is the sum of the similarities to the search request of the document assigned to the information flow,

[0016]

(Equation 2) Can be expressed as

Using the similarity of the information flow, the search results are displayed in order from the one with the largest similarity. There is also a method of displaying information flows having a large number of fresh documents in a higher order, or vice versa, using time information in which each information flow exists.

As another typical method, there is a method of calculating the similarity between an information flow and a search word (search expression), and presenting similar information flows as a search result. Feature vector of information flow to calculate similarity between information flow and search term

[0019]

[Outside 2] Was introduced. Since the components of the information flow are documents, the feature vector of the information flow

[0020]

[Outside 3] Feature vectors of documents that make up the information flow

[0021]

[Outside 4] (Expression (3)) is used.

[0022]

(Equation 3) Here, the feature vector of each search word q is

[0023]

[Outside 5] And Feature vector

[0024]

[Outside 6] Is such that each element q has a one-to-one correspondence with the word q in the text, the value of the element corresponding to the search word q is 1, and the other values are 0. Then, as a similarity between the “information flow” and the search term q, the feature vector of the information flow is

[0025]

[Outside 7] Product (trigonometric function cos in Euclidean space
θ) is used.

Then, an information flow in which the similarity between the feature vector of the information flow and the feature vector of the search word is greater than 0 is extracted, and presented as a search result in descending order of the similarity. By this method, an information flow composed of documents using common words is searched.

The retrieved information flows are topics similar to common words or documents, and these can be compared.

[0028]

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

Referring to FIG. 1, an information flow searching device according to an embodiment of the present invention includes a document input unit 10 and a document storage unit 10.
2, an information power flow extraction unit 103, a search request input unit 104, an information power flow search unit 105, and a display unit 106.

In the information flow presenting apparatus according to the present embodiment, a user inputs a document to be processed by the document input unit 101. The documents to be processed include all the documents input to the computer. For example, there are newspaper articles, HTML files on the Internet, net news, text broadcasting, FM multiplex broadcasting, and broadcast manuscripts on television.

In the document storage unit 102, the document input unit 101
Is stored so that the document of each information transmission medium, or some or all of the information transmission media can be retrieved.

The information flow extraction unit 103 extracts an information flow from a set of documents. For the document set input by the document input unit 101, “1999”, “January 1999”
Classification based on time information such as "January 3, 1999" and topic classification such as "classification category of" traffic accident "" and "classification category of" eight child birth "" are performed. Documents classified into each topic are arranged using time information to generate an information flow.

In the search request input section 104, the user issues a search request by specifying a word to be searched or a document in which information to be searched is written from the extracted information flow.

The information power flow search unit 105 searches for the information power flow extracted by the information power flow extraction unit 103. That is, a document or a classification category that satisfies the search condition input from the search request input unit 104 is extracted from each information flow, the similarity with those search conditions is calculated, and the search results are sequentially sorted from the information flow having the highest similarity. Is displayed on the display unit 106.

The processing flow of the present embodiment will be described using a specific example.

The set of documents to be input to the present apparatus is the same as that of newspaper company A, newspaper company B and newspaper company C since 1988.
Eight-year newspaper article. A document set is input from the document input unit 101. The input document is stored in the document storage unit 102.

Next, the information flow extraction unit 103 extracts an information flow from the input document set. From this result, for example, the information flow of “Typhoon No. 18” in 1996 or 1
The information flow of “Typhoon No. 18” in 997, the information flow of “Japan Professional Soccer League (J-League)” in 1993, the information flow of “J-League” in 1994, and 19
1994 "The World Cup World Cup Information Trends" and 1998 "World Cup France Tournament"
It is assumed that the information tide of information has been extracted (FIG. 3).

Next, the information flow search unit 105 searches for a necessary information flow. For example, suppose we want information on "soccer". When the user inputs “information trend in which a document including the word“ soccer ”exists” as a search request from the search request input unit 104 (step 201), the information of “Japan Professional Soccer League (J-League)” in 1993 Tide "and" Information Tide of "J League""in 1994 and 19
In 1994, the information trend of the “World Cup USA Tournament” and in 1998, the “World Cup France Tournament”
Information flow "includes a document including the word" soccer ", and is extracted as an information flow that meets this condition (FIG. 3).
Steps 202 and 203). The ordering when displaying is
For example, if the ordering condition is "from the latest information flow", the order is displayed in the display unit 105 in order from the newest one (steps 204 and 205).

It is also assumed that information on "typhoon" is desired. When "information flow in which a document including the word" typhoon "exists" is used as the search request (step 201), 1996
The "J League" in 1994, including the article "The Typhoon No. 18 Information Trend", the 1997 "The Typhoon No. 18 Information Trend", and the article "The game was postponed due to the typhoon"
Information flow contains documents containing the word "typhoon"
It is extracted as an information flow that meets this condition (step 2
02, 203). For example, if the ordering condition is “from the largest sum of similarities between words and documents”, the ordering at the time of display does not include much typhoon information.
199 more than the “J-League information trend” in 994
The “information flow of“ Typhoon No. 18 ”” in 6 years and the “information flow of“ Typhoon No. 18 ”” in 1997 are ordered higher and displayed on the display unit 106 (steps 204 and 2).
05).

FIG. 4 shows an information flow presentation device according to another embodiment of the present invention. The information flow presentation device of the present embodiment includes an input device 301, storage devices 302 and 303, an output device 304, a recording medium 305, and a data processing device 306. The input device 301 corresponds to the document input unit 101 in FIG. The storage device 302 corresponds to the document storage unit 102 in FIG. The storage device 303 is a hard disk. The output device 304 corresponds to the display unit 106 in FIG. The recording medium 305 is a recording medium such as an FD (floppy disk), a CD-ROM, and an MO (magneto-optical disk) in which an information flow search program including the processing shown in FIG. 2 is recorded. The data processing device 306 is a CPU that reads the information flow search program from the recording medium 305 into the storage device 303 and executes the program.

[0041]

As described above, according to the present invention, in a large document set, a large amount of information flow generated by using topic information of each document and information of the time when the document was generated is used. Therefore, it is possible to search for only the desired information flow, thereby making it possible to grasp the information desired by the user and the transition of the topic.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a schematic configuration of an information flow search device according to an embodiment of the present invention.

FIG. 2 is an example of a flowchart showing processing from input of a search request to display of a search result in the information flow search device of FIG. 1;

FIG. 3 is a diagram illustrating an example of an output of a search result of the information power flow search device in FIG. 1;

FIG. 4 is a configuration diagram of an information flow search device according to another embodiment of the present invention.

FIG. 5 is a diagram showing an information flow relating to “Typhoon No. 3” obtained by a conventional information flow presentation device.

FIG. 6 is a diagram showing an information flow relating to a “elementary school murder case” obtained by a conventional information flow presentation device.

[Explanation of symbols]

 Reference Signs List 101 Document input unit 102 Document storage unit 103 Information flow extraction unit 104 Search request input unit 105 Information flow search unit 106 Display unit 201 to 205 Step 301 Input device 302, 303 Storage device 304 Output device 305 Recording medium 306 Data processing device

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Daijiro Mori 3-19-2 Nishi-Shinjuku, Shinjuku-ku, Tokyo Japan Telegraph and Telephone Corporation (72) Inventor Kazuo Tanaka 3-192-1, Nishi-Shinjuku, Shinjuku-ku, Tokyo No. Nippon Telegraph and Telephone Corporation F-term (reference) 5B075 ND03 NK06 NK10 NK31 NR12 PQ76 QM08

Claims (3)

[Claims] An information flow search method for extracting a topic and its temporal transition from a large amount of information transmitted from an information transmission medium and searching the document, wherein a document described in a natural language is input. An input step, a document storage step of storing input data obtained in the document input step, and an input data stored in the document storage step.
An information flow extraction step of extracting an information flow, which is a transition of the topic, by categorizing according to the topic and arranging the documents in the order of the time when the document was transmitted; and, from the extracted information flow, a word or a search to be performed. A search request inputting step of designating a document in which information is written and issuing a search request; comparing the information flow with a document including the word specified in the search request inputting step or the specified document, and assigning the information flow An information flow search step of obtaining a similarity to a search request of a document being ordered and ordering information flows from those having the highest similarity or the latest, and a display step of displaying the ordered information flows. Method. 2. An information flow searching apparatus for extracting a topic and its temporal transition from a large amount of information transmitted from an information transmission medium and searching the document, wherein a document described in a natural language is input. An input unit; a document storage unit that stores input data obtained by the document input unit; and input data stored in the document storage unit. An information flow extraction unit for extracting an information flow as a topic transition; and a search request for designating a word or a document in which information to be searched is written from among the extracted information flows. An input unit, comparing the information flow with a document including the word specified in the search request input unit or the specified document, and calculating a similarity to the search request of the document assigned to the information flow, And information tide retrieval unit ordering from those or latest large similarity information tide retrieval device having a display unit that displays the ordered information tide. 3. An information flow search program for extracting and retrieving topics and temporal transitions thereof from a large amount of information transmitted from an information transmission medium, wherein a document described in a natural language is stored in a storage device. By storing the input data stored in the document storage procedure, classifying the input data according to the topic, and arranging the input data in the order in which the documents were transmitted,
An information flow extraction procedure for extracting an information flow that is a transition of a topic, including, from the extracted information flow, a word specified in a search request specifying a word to be searched or a document in which information to be searched is written A document or a specified document is compared with the information flow, a similarity to a search request for a document assigned to the information flow is obtained, and an information flow search for ordering the information flow from the one with the highest similarity or the latest one is performed. A recording medium storing an information flow search program for causing a computer to execute a procedure and a display procedure of displaying the ordered information flow on a display device.