JPH1098655A - Program retrieval device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a program retrieval device with a small calculation amount for the retrieval by which programs such as TV, CTV, teletext and radio programs are retrieved and reservation of programs are made simply and the retrieval is conducted even when a retrieval keyword is not completely accurate. SOLUTION: The device is provided with a database 2 in which program data such as program date and time, channel, program name, contents of program and performers are stored. A word relating to a desired program is entered by an entry means 1, a word dictionary 4 having a characteristic vector of each word is referenced, a characteristic vector corresponding to a question text from the entry means 1 is generated by a vector generating means 5, a retrieval means 6 calculates a distance between the characteristic vector corresponding to the program data in the database with the generated characteristic vector so as to retrieve the desired program.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a program search device, which is used for searching for programs such as television, CATV, teletext, and radio programs, and for video recording reservation.

[0002]

2. Description of the Related Art When watching television, CATV, teletext, and radio programs, and when making a recording reservation for these programs by video, it is necessary to check the program time and channel in newspapers, television information magazines, and the like. In addition, it is necessary to check these information magazines for what programs are being broadcast.

[0003] In recent years, television broadcasting has become more popular than terrestrial broadcasting.
S broadcasts and CS broadcasts can be viewed, and the number of channels has been increased, and various programs can be viewed in CATV and teletext. As described above, when the number of programs increases, it takes time to search for a target program, and it takes time to check an information magazine. Further, when making a reservation for recording these broadcasts on a video, the user inputs the contents manually while looking at the information magazine.

As a conventional technique for searching for such a program (Prior Art 1), a database of video program information is provided as described in JP-A-3-32285, and the data of each program is used for searching. There is a video program search device that can search for a video program by searching for a keyword input by a user by attaching a keyword of the video program.

As a conventional technique for searching for another program (Prior Art 2), as described in JP-A-3-88159, when a program information database is created,
There is a program reservation system that allows a user to specify a genre code when searching for a program by attaching a genre code such as sports or variety to each program, thereby narrowing down the program desired to be reserved.

Further, as a conventional technique for searching for another program (Prior Art 3), as described in Japanese Patent Application Laid-Open No. 7-144633, when searching for a program, the input keyword and the genre name of the program are used. Even if the title name does not match, there is a teletext program search device that enables a search by providing a fuzzy inference program.

[0007]

However, in the prior art 1, since the keyword search is performed, there is a problem that the search cannot be performed unless the input keywords completely match. Further, in the prior art 2, it is not possible to sufficiently narrow down programs by simply setting a genre code, and it is necessary to finely set a genre code in order to sufficiently narrow down a target program. There is a problem that search is difficult.

Further, in the prior art 3, in order to perform fuzzy inference, it is necessary to calculate a fuzzy value for each keyword for all programs in a program database, and the amount of calculation of the fuzzy value becomes enormous. There is. That is, assuming that the number of programs in the program database is n and the number of search keywords is m, n × m fuzzy value calculations are required. Also, every time new program data is added to the database, m fuzzy value calculations are required. Assuming that the number of programs is 1000 and the number of keywords is 10,000, the number of fuzzy value calculations for one program is 100.
00 times, 100,000,000 for the whole database
Times.

[0009] An object of the present invention is to make it easy to search for programs such as television, CATV, teletext, and radio programs and to reserve video recording, and to search even when search keywords do not completely match. It is an object of the present invention to provide a program search device capable of performing a search with a small amount of calculation.

[0010]

According to a first aspect of the present invention, there is provided a program search apparatus for inputting a word associated with a desired program and showing a semantic connection between the word and a characteristic word of the word. A word dictionary stored as a set of feature vectors,
A database stored as a set of a word of program data and a feature vector indicating a semantic connection relationship between the feature word of the word and a database corresponding to the word input from the input means with reference to the word dictionary. Vector generating means for generating a feature vector, search means for calculating a distance between the feature vector generated by the vector generating means and the feature vector in the database, and output means for outputting program data based on the distance. It is characterized by having.

According to a second aspect of the present invention, there is provided the program search apparatus according to the first aspect, wherein the program data is
The system further comprises a data input means comprising a program date, a start time, an end time, a channel, a program name, a program content, etc., and for storing the program data in the database.

According to a third aspect of the present invention, there is provided the program search device according to the first aspect, wherein the program data includes:
It comprises a program date, a start time, an end time, a channel, a program name, a program content, and the like, and is characterized by being used for program setting and changing, or for video recording reservation.

[0013]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The program search device of the present invention utilizes a search device using a feature vector. FIG. 1 shows a configuration of a program search device using the feature vector. This program search apparatus includes an input unit 1 composed of a keyboard, a tablet, and a microphone for a user to input a question sentence that is semantically similar to the content of a desired program, program data and a feature vector corresponding to each program data. Is stored as a set, output means 3 such as a CRT or a printer for outputting a search result, and a word dictionary 4 in which words and feature vectors corresponding to each word are stored as a set When,
A vector generation unit 5 that generates a vector from the word input from the input unit 1 and the word dictionary 4, and a search unit 6 that performs a predetermined operation from the vector generation result from the vector generation unit 5 and the database 2. You. The search device 7 includes a general-purpose computer, a memory, and an external storage device.

First, the feature vector will be described. The feature vector indicates a relationship between a concept and a context of a word in a sentence, and is a vector representation of a degree of a semantic coupling relationship with many feature words. The feature vector of the present invention uses a document search technique based on a context vector in "Associative search from a large-scale database" in the IEICE AI-92-99 (1993-1) issued by IEICE. That is, the “feature vector” in the present embodiment directly corresponds to the above “context vector”.

Assuming that k concept classifications are feature words, the value of each element of the k-dimensional vector corresponds to each feature word. Context vector Xi = (xi) of word i
1, xi2,..., Xik) is 0 ≦ xij
≤ Em. Em is a positive constant. When there is no relation between the word i and the characteristic word j, xij = 0, and when there is a relation, a large value is taken according to the degree of the relation. For example, if the feature vector is composed of five feature words (natural, urban, noise, animal, and green), and the value of each element is binary 0 or 1, the feature of the word "mountain" The vector can be represented as (1,0,0,1,1) and so on.

Next, the procedure of a program search will be described.
From the input means 1, the user inputs a question sentence having a content meaningfully similar to the content of the desired program. The question sentence may be input for each word, sentence, or natural language as long as it includes words. As shown in FIG. 7, if the input question sentence is a sentence or the like, individual words are extracted by the vector generation means 5, a feature vector corresponding to each word is read from the word dictionary 4, Compute the sum and normalize each vector. The search means 6 calculates the distance between the sum of the vectors and the feature vectors corresponding to the respective program data in the database 2, and uses those program data as search results in order from the program data having the closest feature vector. Output to output means 3.

The distance calculation by the search means 6 will be specifically described. In the present invention, inner product calculation is performed for calculating the distance. X = (x1, x2,
, Xm) and Y = (y1, y2,..., Ym)
The inner product XY is expressed as XY = x1 * y1 + x2 * y2 + ... + xm * ym. The larger the inner product value, the closer the distance. For example, the feature vector Q of the question sentence q is Q = (3,5,4,2,4,5,2,1), and the feature vectors S and T of the data s and t in the database 2 are S = (4,5,4,1,4,5,0,1) If T = (5,0,4,6,3,1,3,2), QS = 3 × 4 + 5 × 5 + 4 × 4 + 2 × 1 + 4 × 4 + 5 × 5 + 2 × 0 + 1 × 1 = 97 Q · T = 3 × 5 + 5 × 0 + 4 × 4 + 2 × 6 + 4 × 3 + 5 × 1 + 2 × 3 + 1 × 2 = 68. The distance will be short.

[0018]

FIG. 2 shows a program search apparatus according to an embodiment of the present invention. Here, the same parts as those in FIG. 1 are denoted by the same reference numerals. As compared with FIG. 1, a data input unit 8 for inputting data to the database 2 and a controlled device 1
Wired or wireless communication means 9 for controlling 0 and transmitting program data, and a controlled device 10 are added. The controlled device 10 is a receiver or a receiver such as a television, a CATV receiver, a teletext receiver, a radio, or a video. The search device 7 outputs all or a part of the program data selected by the input unit 1 from the program data candidates of the search result output to the output unit 3 to the controlled device 10 via the communication unit 9.

The search device 7 receives program data such as date, start time, end time, channel, program name, program contents, performers, etc. of each program from the data input means 8, and stores them in the database 2. . FIG. 3 shows a data configuration example of the program data.

Next, the creation of the program database of the database 2 will be described. The input from the data input means 8 includes input from a keyboard, OCR input, voice input, input from online data such as personal computer communication, reception of teletext data, and the like. The input program data is converted into a feature vector for each program data, and the program data and the feature vector are stored in the database 2 as a set.
Here, generation of the feature vector is performed by the vector generation unit 5.
, An individual word is extracted from the program data, a feature vector corresponding to each word is read from the word dictionary 4, the sum of the vectors is calculated, and the vector is normalized so that the size of the vector is constant. Do it by doing.

Specifically, a process of replacing the program data with a feature vector will be described. This process is similar to the process of generating a feature vector from a word. Here, it is assumed that the contents of the program data are shown in FIG. 4 and a part of the contents of the word dictionary 4 is shown in FIG. Here, the program name, program contents, and other data portions are extracted from the program data, and the words are extracted by removing the particles. Here, “news”, “today”, “special feature”, “spring”, and “topic” are extracted. Next, the feature vectors of these words are read from the word dictionary 4, and the sum Vs of these vectors is calculated. If the feature vector of the word X is represented by V (X), Vs = V (news) × 2 + V (today) + V (special feature) +
V (spring) + V (topic). "News" is extracted in two places, so
Has been doubled. When the value of the feature vector in FIG. 5 is substituted, Vs = (3, 5, 0, 4, 2,..., 3). Next, this vector is normalized. Assuming that the normalized vector is Vn, it is calculated as Vn = a × Vs / │Vs│. Here, the magnitude of the vector after a is the normalized, ￨Vs￨ takes a positive value in the magnitude of the vector Vs, Vs = (v1, v2 , ..., vm) and when, ￨Vs￨ ^{2 = | v1 | 2 + | v2} | 2 + ... + | a ² | vm. Here, normalization is performed with a = 10. The feature vector and the program data thus normalized are stored in the database 2 as a set.

FIG. 6 is a flowchart of a program search process. First, a question sentence of a search condition such as a program name or a program content that is semantically similar to the program content to be searched is input from the input unit 1 (S1). The search device 7 searches for program data in the database 2 (S2). Since the search performed here is an associative search, even if there is no completely matching data in the program data, it is possible to search for program data having semantically similar contents. The search result of the search device 7 is output to the output means 3 (S3), and the content of the search is confirmed and selected (S4). The selection is performed by inputting desired program data from among the program data candidates displayed on the CRT, for example.
This is done by specifying. Then, necessary program data among the program data selected from the search results is transmitted to the controlled device 10 via the communication means 9. In the case of setting or changing the channel (S5), the data of the channel in the program data is transmitted to the controlled device 10, and the channel of the controlled device 10 is set or changed based on the received channel. If the program has not started yet, the data of the date, start time and, if necessary, end time in the program data are transmitted and set in the controlled device 10. On the other hand, in the case of the video recording reservation (S5), the controlled device 10
The data of the channel, date, start time and, if necessary, end time in the program data to be reserved for video recording is transmitted, and video recording reservation of the controlled device 10 is set based on the received data. By outputting the program data to the output means 3 at the time of the video recording reservation, the program contents can be confirmed, so that it is possible to confirm whether the video recording reservation contents are correct.

Since the search device 7 performs a search using a feature vector, it is possible to search for a program by performing a semantic search. For example, when "comedy program" is inputted from the input means 1, "comedy", "comic", "comedy",
Programs including data such as "rakugo" can also be searched. As described above, it is possible to search for a program having a semantically similar content without specifically inputting the content of the target program. Further, by adding conditions such as date, day of the week, time, performer, channel, and the like, the search results can be narrowed down, and a target program can be found.

Also, the amount of calculation for the associative search is smaller than that using fuzzy inference. To rewrite the program data in the database to the feature vector,
For one program, the feature vector of the word included in the program data is read from the word dictionary, and the feature vector for the program data is calculated.
The number of times of calculation of the feature vector is one, and the number of programs in the database is n, where n is the number of programs in the database. Assuming that the number of programs is 1000, the number of feature vector calculations for the entire database is 1000, which is irrelevant to the number of words in the word dictionary.

[0025]

According to the program search apparatus of the present invention, a semantic search using a feature vector is performed in a program search, so that a semantic search can be performed without specifically specifying the contents of a target program. You can easily search for the desired program by entering a content similar to that in the case of program setting, ensuring program settings and changes in the case of program settings, and preventing mistakes in the case of video recording reservation .

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a program search device of the present invention.

FIG. 2 is a configuration diagram of an embodiment of the present invention.

FIG. 3 is a configuration diagram of program data according to the embodiment of the present invention.

FIG. 4 is a specific example of program data according to the embodiment of the present invention.

FIG. 5 is a diagram showing a part of the word dictionary according to the embodiment of the present invention.

FIG. 6 is a flowchart of a search according to the embodiment of the present invention.

FIG. 7 is a flowchart of feature vectorization according to the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Input means 2 Database 3 Output means 4 Word dictionary 5 Vector generation means 6 Search means

Claims (3)

[Claims] An input means for inputting a word related to a desired program, a word dictionary stored as a set of a word and a feature vector indicating a semantic connection relationship between the word and a feature word of the word, A database stored as a set of a word of data and a feature vector indicating a semantic connection relationship between the feature word of the word and a feature corresponding to the word input from the input means by referring to the word dictionary Vector generating means for generating a vector; searching means for calculating a distance between the feature vector generated by the vector generating means and the feature vector in the database; and output means for outputting program data based on the distance. A program search device characterized in that: 2. The program data comprises a program date, a start time, an end time, a channel, a program name, a program content, and the like, and further comprising data input means for storing the program data in the database. Claim 1
2. The program search device according to 1. 3. The program data includes a program date, a start time, an end time, a channel, a program name, a program content, and the like, and is used for program setting and changing, or for video recording reservation. The program search device according to claim 1.