JP2022026752A - Knowledge database generation device, program, and automated answering system - Google Patents

Abstract

To provide a knowledge database generation device, a program, and an automated answering system, which can easily construct a knowledge database having highly accurate data.SOLUTION: A knowledge DB generation server 1 includes: a character processing section 12 that recognizes or acquires a character string included in data to be registered and extracts a word included in the character string; an image processing section 14 that performs image recognition processing on the data to be registered and acquires the recognized word; a word classification section 16 that acquires classification of the word obtained as a result of performing processing of the character processing section 12 and the image processing section 14 using dictionary data; and a data registration section 18 that associates the data to be registered with the word and the classification and stores in a knowledge DB 3.SELECTED DRAWING: Figure 1

Description

The present invention relates to a knowledge database generator, a program and an automated response system.

Conventionally, from the viewpoint of improving work efficiency, an effort has been made to install a display and a microphone at a reception desk of a company, etc., and to perform customer service work such as window work by the displayed virtual character or the like. In such a system, the user's question is recognized by voice recognition via a microphone, the answer is searched from the knowledge database, and the user's question is answered by displaying it on the display and / or speaking from the speaker. .. As an example of such a dialogue system, for example, the technique described in Patent Document 1 is disclosed.

Japanese Unexamined Patent Publication No. 2007-148118

In order to realize the dialogue system as described in Patent Document 1, it is necessary to register the data in the knowledge database in advance, but in that case, the amount of data to be registered in the knowledge database and the accuracy of the data are important. Therefore, it is desired to more easily construct a knowledge database containing highly accurate data.

Therefore, an object of the present invention is to provide a knowledge database generator, a program, and an automatic response system capable of easily constructing a knowledge database having highly accurate data.

The present invention solves the above-mentioned problems by the following solution means.
The first invention is a character processing means for recognizing or acquiring a character string included in the registration target data and extracting a word included in the character string, and an image recognition process for the registration target data to recognize the word. The word classification means for acquiring the word classification obtained as a result of processing the character processing means and the image processing means, and the word classification means for acquiring the word classification using dictionary data, and the registration target data. , A knowledge database generation device comprising a data registration means for associating the word with the classification and storing it in the knowledge database.
The second invention is the knowledge database generation device of the first invention, wherein the image recognition process is a knowledge database generation device that recognizes an image scene.
According to the third invention, in the knowledge database generator of the first invention or the second invention, the registration target data is analyzed to confirm whether or not the image and the character string overlap, and if there is an overlap, the third invention is performed. An image processing means for deleting a character string area including the character string and complementing the deleted character string area with an image around the character string area is provided, and the image processing means is based on the image processing means. It is a knowledge database generation device that performs processing using the registered data after processing.
The fourth invention is a hypernym of the word obtained as a result of processing the character processing means and the image processing means in any of the knowledge database generators from the first invention to the third invention. The word classification means is a knowledge database generation device that further acquires the classification by using the hypernym acquired by the hypernym acquisition means as the word.
In the fifth invention, the weight of the word obtained as a result of processing the character processing means and the image processing means in any of the knowledge database generation devices from the first invention to the fourth invention is used. The data registration means is a knowledge database generation device including a weight determining means for determining, which further associates the weight determined by the weight determining means with the word and stores the weight in the knowledge database.
A sixth aspect of the invention is the knowledge database generation device of the fifth aspect, wherein the weight determination means includes the frequency of appearance of the word in the registration target data, the mode of highlighting the word, and the area of the registration target data. A knowledge database generator that determines the weight of the word based on at least one of the proportions of the image corresponding to the word in.
The seventh invention comprises a color information acquisition means for analyzing the registration target data and acquiring color information in any of the knowledge database generation devices from the first invention to the sixth invention, and the data registration means. Is a knowledge database generation device that further associates the color information acquired by the color information acquisition means with the registration target data and stores it in the knowledge database.
According to the eighth aspect of the invention, in any one of the knowledge database generators from the first invention to the seventh invention, the character processing means extracts the character string to be a heading and has the same or similar heading. It is a knowledge database generation device provided with an association means for associating the registration target data.
The ninth invention is a program for making a computer function as a knowledge database generator according to any one of the first invention to the eighth invention.
The tenth invention is an automatic response system using the knowledge database generated by the knowledge database generator of the fifth invention or the sixth invention, and is received by a question receiving means for receiving a question and a question receiving means. The question was obtained by searching the knowledge database based on the question analysis means for performing morphological analysis on the question in character string and the analysis result by the question analysis means, and acquiring the question as a search result. It is an automatic response system including a response output means for outputting the registration target data, giving priority to a word having a large total weight.

According to the present invention, it is possible to provide a knowledge database generator, a program, and an automatic response system capable of easily constructing a knowledge database having highly accurate data.

It is a functional block diagram of the automatic response system which concerns on this embodiment. It is a flowchart which shows the knowledge data registration process in the knowledge DB generation server which concerns on this embodiment. It is a figure which shows the example of the registration target data to be registered in the knowledge DB which concerns on this embodiment. It is a flowchart which shows the character processing in the knowledge DB generation server which concerns on this embodiment. It is a figure which shows the specific example of the image processing processing in the character processing which concerns on this embodiment. It is a flowchart which shows the image processing in the knowledge DB generation server which concerns on this embodiment. It is a figure which shows the example of the knowledge data which is registered in the knowledge DB generated by the knowledge DB generation server which concerns on this embodiment. It is a flowchart which shows the question answering process in the automatic answering apparatus which concerns on this embodiment. It is a figure which shows the specific example in the automatic response apparatus which concerns on this embodiment.

Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings. It should be noted that this is only an example, and the technical scope of the present invention is not limited to this.
(Embodiment)
<Automatic response system 100>
FIG. 1 is a functional block diagram of the automatic response system 100 according to the present embodiment.
The automatic response system 100 is a system for responding to a question received by the automatic response device 4 by using the knowledge DB (database) 3. Further, in the automatic response system 100, the knowledge DB generation server 1 generates the knowledge DB 3 by using an existing medium such as a leaflet or a pamphlet.
The automatic response system 100 includes a knowledge DB generation server 1, a knowledge DB 3, and an automatic response device 4.

<Knowledge DB generation server 1>
The knowledge DB generation server 1 is a device that generates the knowledge DB 3 by registering the registration target data and various information obtained from the registration target data in the knowledge DB 3.
The knowledge DB generation server 1 is, for example, a server, a personal computer (PC), or the like.
The knowledge DB generation server 1 includes a control unit 10, a storage unit 20, and a communication interface unit 29.

The control unit 10 is a CPU (central processing unit) that controls the entire knowledge DB generation server 1. The control unit 10 appropriately reads out and executes an OS (operating system) and an application program stored in the storage unit 20 to cooperate with the above-mentioned hardware and execute various functions.
The control unit 10 includes a target data receiving unit 11, a character processing unit 12 (character processing means), a target data confirmation processing unit 13 (image processing means, color information acquisition means), and an image processing unit 14 (image processing means). The upper word acquisition unit 15 (upper word acquisition means), the word classification unit 16 (word classification means), the weight determination unit 17 (weight determination means), and the data registration unit 18 (data registration means) are provided.

The target data receiving unit 11 receives the registration target data. Here, the registration target data is the knowledge data to be registered in the knowledge DB 3. The data to be registered may be, for example, various existing media, and may be distributed in paper media such as leaflets and pamphlets, as well as digitized data such as Web pages and SNS (Social Networking Services). It may be present, or it may be a handwritten memo or the like. In the present embodiment, the data to be registered will be described as including characters and images, such as a leaflet or a pamphlet. Actually, the data to be registered may include only characters or only images.
When the target data receiving unit 11 is a paper medium, the target data receiving unit 11 is imaged by, for example, a scanner, a camera, or the like (not shown), and receives the imaged image data as the registration target data.
Further, the target data receiving unit 11 may collectively accept a plurality of pages such as pamphlets as one registration target data, or may accept each page as registration target data.

The character processing unit 12 processes the text (character string) included in the received registration target data.
The character processing unit 12 acquires the text included in the registration target data, or recognizes it by the character recognition process and acquires the word included in the recognized text. The character processing unit 12 can perform a process of acquiring a word from a text as a natural language process, for example, by morphological analysis or the like.

The target data confirmation processing unit 13 analyzes the registration target data and confirms whether or not the image and the text overlap. Then, when there is an overlap, that is, when the text overlaps the image, the target data confirmation processing unit 13 deletes the text area including the text, and the deleted text area is used as the image around the text area. Performs complementary image processing processing using.
Further, the target data confirmation processing unit 13 acquires the color information of the registration target data. The target data confirmation processing unit 13 acquires color information by analyzing the hue of the entire image of the registration target data. The color information is acquired, for example, as a color code in which the RGB value is represented by a hexadecimal number.

The image processing unit 14 performs image recognition processing on the registration target data and acquires the recognized word. More specifically, the image processing unit 14 detects and analyzes each image included, and acquires a word related to each image, for example. Further, for example, the image processing unit 14 utilizes an existing image analysis technique to detect and cut out each image included in the registration target data, and labels each image to label the image. Get as. The image processing unit 14 uses an image DB (not shown) in which an image and a word are associated with each image for labeling, and acquires a word associated with an image in an image DB similar to each image. You may. Further, the image processing unit 14 recognizes the scene of the image. Here, when the image processing unit 14 performs the image processing processing by the target data confirmation processing unit 13, the image processing unit 14 may detect each image from the processed target data which is the registration target data after the image processing processing. good.
If the registration target data itself is labeled in advance, the image processing unit 14 may acquire the label.

The hypernym acquisition unit 15 acquires the hypernym of the word acquired by the character processing unit 12 and the image processing unit 14. The hypernym acquisition unit 15 acquires a hypernym corresponding to a hypernym of a word by using, for example, a term dictionary (not shown) in which the word is associated with the hypernym. The hypernyms are, for example, "fresh fish" and "food" when the word is "tuna".
The word classification unit 16 acquires the classification of words (including hypernyms). The word classification unit 16 refers to the scene-related term dictionary 22 (dictionary data) and classifies words into "objects" and "scenes". As a result, the word classification unit 16 may further categorize the words classified as "objects".

The weight determination unit 17 determines the weight of each word. Here, the weight determination unit 17 may increase the weight of the word acquired from both the character and the image, for example. Further, the weight determination unit 17 may determine the weight according to the size of the character and the font (mode), and the weight of the word having a larger character size and being bold or underlined is made larger. May be good. Further, the weight determination unit 17 may determine the weight according to the size of the image. For example, the larger the ratio of the size (area) of the image to the entire area of the registration target data, the larger the weight. May be good.

The data registration unit 18 associates a word with a classification with the data to be registered and registers it in the knowledge DB 3. The data registration unit 18 may further associate the color information acquired by the target data confirmation processing unit 13 with the registration target data. Further, the data registration unit 18 may further associate the weight determined by the weight determination unit 17 with the word.

The storage unit 20 is a storage area for a hard disk, a semiconductor memory element, or the like for storing programs, data, and the like necessary for the control unit 10 to execute various processes.
The storage unit 20 includes a program storage unit 21 and a scene-related term dictionary 22.
The program storage unit 21 is a storage area for storing a program. The program storage unit 21 stores the program 21a.
The program 21a is a program for executing various functions of the control unit 10.

The scene-related term dictionary 22 is a dictionary that stores at least words related to the scene. A scene means a scene or a scene, and is a word that expresses a state as a whole that is different from individual objects. The scene-related term dictionary 22 may associate a word with either a scene or an object. The scene-related term dictionary 22 is prepared in advance.
The communication interface unit 29 is an interface for communicating with the knowledge DB 3, other DBs, other devices, and the like via the communication network.

<Knowledge DB3>
The knowledge DB 3 is a database that stores information on registration target data including words and classifications acquired from the registration target data by the knowledge DB generation server 1. In FIG. 1, the knowledge DB 3 is provided outside the knowledge DB generation server 1. The knowledge DB 3 has a large storage area for storing a large amount of information including a large amount of image data.

<Automatic response device 4>
The automatic response device 4 is a device that answers a question by using the knowledge DB 3.
The automatic response device 4 is, for example, a PC, a smartphone, or the like, and may be a robot or the like. In the case of a PC, a robot, or the like, the automatic response device 4 is provided at a tourist information center, reception desks, etc. at various places, and responds to questions from ordinary people who visit the place where the automatic response device 4 is installed. It is the answer. Further, in the case of a PC, a smartphone, or the like, the automatic response device 4 answers, for example, a question of the owner of the automatic response device 4. In the following description, a person who asks a question using the automatic response device 4 is referred to as a user.

The automatic response device 4 includes a control unit 40, a storage unit 50, an input unit 56, a display unit 57, and a communication interface unit 59.
The control unit 40 is a CPU that controls the entire automatic response device 4. The control unit 40 appropriately reads and executes the OS and application programs stored in the storage unit 50, thereby collaborating with the above-mentioned hardware to execute various functions.
The control unit 40 includes a reception processing unit 41 (question receiving means, question analysis means), a DB processing unit 42 (answer output means), and an answer output unit 43 (answer output means).

The reception processing unit 41 receives questions from users. When the user inputs characters using a keyboard or the like or speaks to a microphone, the reception processing unit 41 receives a question from the user.
Then, when the received question needs to be converted into text, the reception processing unit 41 performs morphological analysis after converting it into text, and extracts, for example, words included in the question.

The DB processing unit 42 searches the knowledge DB 3 based on the extracted words. Then, the DB processing unit 42 generates a list of the results of searching the knowledge DB 3. Here, the DB processing unit 42 may generate a result list in which the registration target data including a word having a large weight is ranked higher.
The answer output unit 43 causes the display unit 57 to output higher-level registration target data from the result list generated by the DB processing unit 42.

The storage unit 50 is a storage area for a hard disk, a semiconductor memory element, or the like for storing programs, data, and the like necessary for the control unit 40 to execute various processes.
The input unit 56 is, for example, an input device such as a keyboard or a mouse. The input unit 56 may be, for example, a voice input device such as a microphone.
The display unit 57 is, for example, a display device such as an LCD (Liquid Crystal Display).
The communication interface unit 59 is an interface for communicating with the knowledge DB 3 and the like via the communication network.

Here, the computer means an information processing device provided with a control unit, a storage device, and the like, and the knowledge DB generation server 1 and the automatic response device 4 are information processing devices provided with a control unit, a storage unit, and the like, respectively. Included in the concept of computers.

<Explanation of processing>
Next, the process of registering the knowledge data in the knowledge DB 3 will be described.
FIG. 2 is a flowchart showing a knowledge data registration process in the knowledge DB generation server 1 according to the present embodiment.
FIG. 3 is a diagram showing an example of registration target data to be registered in the knowledge DB 3 according to the present embodiment.
FIG. 4 is a flowchart showing character processing in the knowledge DB generation server 1 according to the present embodiment.
FIG. 5 is a diagram showing a specific example of image processing processing in the character processing according to the present embodiment.
FIG. 6 is a flowchart showing image processing in the knowledge DB generation server 1 according to the present embodiment.
FIG. 7 is a diagram showing an example of knowledge data 31 registered in the knowledge DB 3 generated by the knowledge DB generation server 1 according to the present embodiment.

The knowledge data registration process shown in FIG. 2 is a process performed for each registration target data. When there are a plurality of registration target data, the control unit 10 repeats the knowledge data registration process a plurality of times for each registration target data.
In step S (hereinafter, simply referred to as “S”) 11 of FIG. 2, the control unit 10 (target data reception unit 11) of the knowledge DB generation server 1 receives the registration target data. Here, if the media to be registered is not digitized, the control unit 10 uses a scanner or the like to image the media, and then accepts the imaged data as the data to be registered.
As an example of the data to be registered, the imaged pamphlet 61 is shown in FIG.
The pamphlet 61 shown in FIG. 3 relates to gourmet food. The pamphlet 61 includes texts 62a to 62c and images 63a to 63c.

In S12 of FIG. 2, the control unit 10 performs character processing.
Here, the character processing will be described with reference to FIG.
In S21 of FIG. 4, the control unit 10 (character processing unit 12) recognizes the text included in the registration target data by the character recognition process. In the example of the pamphlet 61 of FIG. 3, the text 62a of "delicious gourmet MAP", the text 62b of "sea bream", and the text 62c of "tuna" are recognized. The control unit 10 may recognize the text 62a as three characters, "delicious", "gourmand", and "MAP", as separate texts for each line.
In S22 of FIG. 4, the control unit 10 (character processing unit 12) acquires a word included in the text. Then, the control unit 10 temporarily stores the acquired word in the storage unit 20.

In S23, the control unit 10 (target data confirmation processing unit 13) confirms the overlap between the image and the text. For example, the texts 62b and 62c of FIG. 3 overlap with the images 63b and 63c, respectively.
In S24, the control unit 10 determines whether or not there is an overlap between the image and the text. If there is even one overlap between the image and the text (S24: YES), the control unit 10 shifts the processing to S25. On the other hand, when there is no overlap between the image and the text (S24: NO), the control unit 10 shifts the processing to S13 in FIG.
In S25, the control unit 10 (target data confirmation processing unit 13) performs image processing processing on the overlapping portion of the image and the text of the registration target data, and generates the processed target data. Then, the control unit 10 temporarily stores the processed target data after the image processing process in the storage unit 20. After that, the control unit 10 shifts the processing to S13 in FIG.

Here, the image processing process will be described with reference to FIG.
FIG. 5A shows an image 63c of the pamphlet 61 including a text 62c having an overlapping portion between the image and the text. The image processing process for the image 63c is performed through the following steps.
The control unit 10 outlines the text area including the text 62c with respect to the image 63c. FIG. 5B is an image 70 after the whitening process by the control unit 10, and includes the whitening unit 71.
Next, the control unit 10 performs a process of complementing the white portion 71 of the image 70 with the average color around the white portion 71, and includes the complementary image 73 as shown in FIG. 5 (C). The processed image 72 is generated.
The control unit 10 also performs image processing processing on the image 63b including the text 62b of the pamphlet 61. By doing so, the processed target data after the image processing process has no overlap between the image and the text.

In S13 of FIG. 2, the control unit 10 performs image processing.
Here, image processing will be described with reference to FIG.
In S31 of FIG. 6, the control unit 10 (target data confirmation processing unit 13) acquires the color information of the registration target data (when image processing is performed, the processed target data). At that time, the control unit 10 acquires the ratio of each color information in the entire registration target data. Since the color information expresses the impression of the entire color of the data to be registered, it can be used in a vague similar image search. The control unit 10 temporarily stores the acquired color information in the storage unit 20.

In S32, the control unit 10 (image processing unit 14) performs image cutting processing for detecting and cutting out each image included in the registration target data. By this processing, three images from images 63a to 63c shown in FIG. 3 can be cut out. Since the control unit 10 cuts out an image from the processed target data after the image processing process, the cutting accuracy of the image can be made highly accurate.

In S33, the control unit 10 (image processing unit 14) performs image recognition processing and acquires the recognized word. The control unit 10 may acquire a word related to the cut out image by using the image DB (not shown). Words obtained from an image include not only those related to objects contained in the image but also those related to the scene of the entire image. The image used in this process does not contain text. Therefore, the process of acquiring a word from an image can be performed with high accuracy. The control unit 10 temporarily stores the acquired word in the storage unit 20. After that, the control unit 10 shifts the processing to S14 in FIG.

In S14 of FIG. 2, the control unit 10 (hypernym acquisition unit 15) acquires the hypernym of the word acquired from the text or the image. Here, the control unit 10 may acquire the hypernyms of all the words and associate the acquired hypernyms with the words. Further, the control unit 10 does not need to acquire the hypernyms of all the words, and may acquire the hypernyms only when, for example, there is a common hypernym in a plurality of words. The control unit 10 temporarily stores the acquired hypernym in the storage unit 20.

In S15, the control unit 10 (word classification unit 16) obtains the classification of each word (including the hypernym) by referring to the scene-related term dictionary 22. Then, the control unit 10 associates each word with the acquired classification.
In S16, the control unit 10 (weight determination unit 17) determines the weight of each word. As described above, the control unit 10 can use various indexes based on text and images as the weight of each word. Further, the control unit 10 may make the weight of the hypernym the same as the weight of the corresponding word, or may make the weight of the hypernym of a plurality of words the same as the weight of the large word. ..

In S17, the control unit 10 (data registration unit 18) associates the word, the classification, and the weight with the data to be registered, further associates the color information, and registers the data in the knowledge DB 3. Here, when the control unit 10 acquires the hypernyms of all the words, only the hypernyms associated with the words having a weight of a predetermined weight or more may be registered.
After that, the control unit 10 ends this process.

FIG. 7 shows an example of the knowledge data 31 registered in the knowledge DB 3.
The knowledge data 31 shown in FIG. 7 is the data to be registered in the pamphlet 61 shown in FIG.
The knowledge data 31 includes a data unit 31a, a keyword classification unit 31b, and a color information unit 31c.
The data unit 31a stores the registration target data and, if generated, the processed target data.
The keyword classification unit 31b stores words, classifications, and weights in association with each other.
For example, when the word is "tuna", it is a word extracted from text and an image, the classification is "food" of "object", and the weight is "2".
In the color information unit 31c, the color information of each color constituting the entire pamphlet 61 and the ratio thereof are stored in association with each other.

By this knowledge data registration process, the knowledge DB generation server 1 registers the registration target data in the knowledge DB 3 as knowledge data. A plurality of knowledge data 31 corresponding to such registration target data are stored in the knowledge DB 3. Then, the knowledge DB 3 includes not only the text included in the registration target data but also the words related to the image. And the word about the image includes not only the object contained in the image but also the scene obtained from the image. Therefore, the knowledge data registered in the knowledge DB 3 can be converted into highly accurate data that can be used at the time of search.

Next, the process related to the automatic response using the knowledge DB 3 will be described.
FIG. 8 is a flowchart showing a question answering process in the automatic answering device 4 according to the present embodiment.
FIG. 9 is a diagram showing a specific example of the automatic response device 4 according to the present embodiment.
In S41 of FIG. 8, the control unit 40 (reception processing unit 41) of the automatic response device 4 receives a question from the user via the input unit 56.
In S42, the control unit 40 (reception processing unit 41) converts the received question into a text. When a question from a user is received using an input device such as a keyboard, the process is not necessary because it has already been converted into text.
In S43, the control unit 40 (reception processing unit 41) performs morphological analysis processing and extracts words.

In S44, the control unit 40 (DB processing unit 42) searches the knowledge DB 3 based on the extracted words. Then, the control unit 40 (DB processing unit 42) generates a list of search results in which the registration target data including a word having a large weight is ranked higher.
In S45, the control unit 40 (answer output unit 43) outputs the higher-level registration target data from the generated search result list to the display unit 57. After that, the control unit 40 ends this process.

FIG. 9 shows an example of a usage scene by a user.
When the user says, "I want to find a restaurant where I can eat delicious rice that can be used on a date at night.", The control unit 40 of the automatic response device 4 asks a question via the microphone 56a (input unit 56). Accept.
Then, as a result of processing by the automatic response device 4, the control unit 40 displays the pamphlet 61 on the display 57a (display unit 57), and so that the virtual character speaks, "I have this information, how about it?" Is output from the speaker 58. At that time, the control unit 40 may highlight the portion of the pamphlet 61 that is related to the keyword statement by highlighting or the like.

Here, when the user further requests other information, the control unit 40 can perform an operation by saying "What else?" Or by performing an operation by an input unit (not shown) indicating the following. The registration target data may be output to the display unit 57 in order from the list of search results in which the registration target data including a word having a large weight is ranked higher.

As described above, the automatic response system 100 of the present embodiment has the following effects.
(1) The knowledge DB generation server 1 acquires the classification of words obtained from the entire registration target data and the texts and images contained in the registration target data, and associates the words and classifications with the registration target data in the knowledge DB 3. to register. Therefore, various words obtained from the registration target data can be associated with the classification and registered as knowledge data in the knowledge DB 3.
(2) The knowledge DB generation server 1 refers to the scene-related term dictionary 22 that stores terms related to the scene, and classifies the words as at least whether or not they are scenes. Therefore, each word can be classified as the whole of the data to be registered or as a scene that conveys the atmosphere of the image.
(3) The knowledge DB generation server 1 confirms whether or not the image and the text overlap, and if they overlap, the text is deleted from the image, and the processed target data including the processed image is subjected to processing. Image processing is performed. Therefore, the accuracy of image processing can be improved by targeting an image that does not contain text as a processing target.

(4) Since the knowledge DB generation server 1 acquires the hypernym of a word and acquires the acquired hypernym classification, the acquired hypernym and classification can be registered in the knowledge DB 3 and the knowledge to be registered in the knowledge DB 3. The data can be made more accurate.
(5) The knowledge DB generation server 1 determines the weight of a word, and further associates the weight with the word and registers the weight in the knowledge DB 3. The weight is determined by the frequency of appearance of words, the degree of emphasis, and the degree of occupancy of the image. Therefore, when the knowledge DB 3 is used for answering a question, there is a possibility that a better answer with higher accuracy can be output by using the weight.

(6) When the automatic response device 4 receives a question, the automatic response device 4 converts the question into text and performs morphological analysis, searches for the knowledge DB 3 generated by the knowledge DB generation server 1 based on the analysis result, and sequentially in descending order of weight. Output the registration target data as an answer to the question. Therefore, it is possible to answer the question using the knowledge DB3. At that time, by using the weight, the registration target data of the knowledge DB 3 more suitable for the question can be output as an answer.

Although the embodiments of the present invention have been described above, the present invention is not limited to the above-described embodiments. Moreover, the effects described in the embodiments are merely a list of the most suitable effects resulting from the present invention, and the effects according to the present invention are not limited to those described in the embodiments. The above-described embodiment and the modified form described later can be used in combination as appropriate, but detailed description thereof will be omitted.

(Transformed form)
(1) In the present embodiment, the knowledge DB generation server and the automatic response device have been described as different devices, but the present invention is not limited to this. It may be a device in which the functions of the knowledge DB generation server and the automatic response device are integrated. Further, the knowledge DB may be stored in the storage unit of the knowledge DB generation server, for example.

(2) In the present embodiment, the determination of the weight is described by taking as an example a word in which a word appearing a plurality of times is considered to be an important word and the weight is increased, but the weight is not limited to this. For example, the larger the occupancy ratio of the image representing the word in the registration target data, the larger the weight may be. Further, when the font of the character is large or the character is bold, the weight may be increased.

(3) In the present embodiment, in the use of the knowledge DB, the data to be output in order from the registration target data having the largest weight has been described as an example, but the present invention is not limited to this. For example, a plurality of registration target data may be output as a list as a search result, and when the user specifies the registration target data from the list, the details of the designated registration target data may be output.
(4) The color information registered in the knowledge DB according to the present embodiment can be utilized, for example, by a vague similar image search. Also, it can be used as key information when you want to obtain the same information again after searching once.
(5) The knowledge data generated in the present embodiment may be further associated with each other. For example, the character processing means extracts the text to be the heading. The text to be the heading may be specified in advance. Then, the association means may associate the registration target data having the same or similar headings.

1 Knowledge DB generation server 3 Knowledge DB
4 Automatic response device 10, 40 Control unit 12 Character processing unit 13 Target data confirmation processing unit 14 Image processing unit 15 Hypernym acquisition unit 16 Word classification unit 17 Weight determination unit 18 Data registration unit 20, 50 Storage unit 21a Program 22 Scene-related Term dictionary 41 Reception processing unit 42 DB processing unit 43 Answer output unit 56 Input unit 57 Display unit 61 Brochure 100 Automatic response system

Claims (10)

A character processing means for recognizing or acquiring a character string included in the data to be registered and extracting a word contained in the character string, and a character processing means.
An image processing means that performs image recognition processing on the registration target data and acquires the recognized word,
A word classification means for acquiring the classification of the word obtained as a result of processing the character processing means and the image processing means using dictionary data, and a word classification means.
A data registration means for associating the word with the classification with the registration target data and storing the word in the knowledge database.
A knowledge database generator. In the knowledge database generator according to claim 1,
The image recognition process is a knowledge database generation device that recognizes an image scene. In the knowledge database generator according to claim 1 or 2.
The registration target data is analyzed to confirm whether or not the image and the character string overlap, and if there is an overlap, the character string area including the character string is deleted, and the deleted character string area is used as the character. Equipped with image processing means to complement using the image around the column area,
The image processing means is a knowledge database generation device that performs processing using the registration target data after processing by the image processing means. In the knowledge database generator according to any one of claims 1 to 3.
It is provided with a hypernym acquisition means for acquiring a hypernym of the word obtained as a result of processing the character processing means and the image processing means.
The word classification means is a knowledge database generation device that further acquires the classification by using the hypernym acquired by the hypernym acquisition means as the word. In the knowledge database generator according to any one of claims 1 to 4.
A weight determining means for determining the weight of the word obtained as a result of processing the character processing means and the image processing means is provided.
The data registration means is a knowledge database generation device that further associates the word with the weight determined by the weight determination means and stores it in the knowledge database. In the knowledge database generator according to claim 5,
The weight determining means is based on at least one of the frequency of appearance of the word in the registration target data, the mode of highlighting the word, and the proportion of the image corresponding to the word in the region of the registration target data. A knowledge database generator that determines the weight of the word. In the knowledge database generator according to any one of claims 1 to 6.
A color information acquisition means for analyzing the registration target data and acquiring color information is provided.
The data registration means is a knowledge database generation device that further associates the color information acquired by the color information acquisition means with the registration target data and stores it in the knowledge database. In the knowledge database generator according to any one of claims 1 to 7.
The character processing means extracts the character string to be a heading and obtains the character string.
A knowledge database generator comprising associating means for associating the registration target data having the same or similar headings. The program for operating a computer as the knowledge database generator according to any one of claims 1 to 8. An automated response system using the knowledge database generated by the knowledge database generator according to claim 5 or 6.
Question reception means for accepting questions and
A question analysis means for performing morphological analysis on the question received by the question receiving means and converted into a character string, and a question analysis means.
An answer output means that searches the knowledge database based on the analysis result by the question analysis means and outputs the registration target data with priority given to the one having a large total of the weights corresponding to the words acquired as the search result. ,
Equipped with an automatic response system.

Images