Text generation device, text generation method, and text generation program

Abstract

To generate a narration text for a person that is appropriate for a situation.SOLUTION: A text generation device includes a recording unit, a basic information acquisition unit, a word acquisition unit, and a text selection unit. The recording unit records, for each text category that makes up a narration text for a person, a text that is associated with predetermined attribute information and information related to a tag, which is a set of words corresponding to the text. The basic information acquisition unit acquires predetermined basic information about the person including information that can be used to obtain attribute information. The word acquisition unit acquires a set of words that indicate features of a target person who is a subject of a narration text for a person to be generated. The text selection unit selects a predetermined number of texts from texts recorded by the recording unit and sets it to be the selected texts on the basis of the similarity between a set of words contained in a tag and a set of words acquired by the word acquisition unit, and attribute information.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a text generation device, a text generation method, and a text generation program for generating narration text for introducing a person. [Background technology]

[0002] Patent Documents 1 to 3 are known as conventional technologies for generating text. Patent Document 1 states as its problem "To provide an automatic advertising copy generation system capable of generating text with high quality as an advertisement." Patent Document 2 states as its problem "To generate a document suitable for the user by editing a standard text provided to the user using personal information including the user's own profile and hobbies and preferences." Patent Document 3 states as its problem "To provide a document creation system, document creation method, and document creation program that combine abstract and concrete phenomena contained in a text to create a summary text." [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2021-140228 A [Patent Document 2] Patent Publication No. 2022-19266 [Patent Document 3] JP 2023-88336 A Summary of the Invention [Problem to be solved by the invention]

[0004] Patent Document 2 is a conventional technology for generating text related to a person. However, Patent Document 2 is a technology for generating text that is premised on general expressions. The use of narration text for a person may include narration text for a funeral to introduce the deceased, and accurate expressions are not necessarily required, and "appropriate" and "optimal" change depending on the situation. For example, appropriate expressions are thought to be different for narration text for a person at a wedding and for narration text for a person at a funeral. In particular, at a funeral, there is a risk that the sadness of the bereaved family will turn to anger, so narration text for a person needs to be created carefully, and the creator is required to have high skills and experience. However, a funeral is an event that occurs suddenly, and it is difficult to secure someone who can create narration text for a person with expressions appropriate for a funeral in a short period of time and create appropriate text. The present invention has been made in view of the above circumstances, and has an object to provide a text generation device that generates narration text for a character that is appropriately expressed in consideration of the situation. [Means for solving the problem]

[0005] The text generation device of the present invention generates narration text for a person to introduce a person. The text generation device includes a recording unit, a basic information acquisition unit, a word acquisition unit, and a text selection unit. The recording unit records, for each text category constituting the narration text for a person, text associated with predetermined attribute information and information on tags, which are sets of words corresponding to the text, in association with each other. The basic information acquisition unit acquires predetermined basic information on a person including information from which attribute information can be obtained. The word acquisition unit acquires a set of words indicating characteristics of a target person of the narration text for a person to be generated. The text selection unit selects a predetermined number of sentences from the sentences recorded by the recording unit based on the attribute information and the similarity between the set of words included in the tag and the set of words acquired by the word acquisition unit. The selected sentences are selected from the sentences recorded by the recording unit. Effect of the Invention

[0006] According to the text generation device of the present invention, for each text category constituting a narration text for a person, a text associated with attribute information and information on a tag, which is a set of words corresponding to the text, are associated and recorded in a recording unit. Then, a text is selected based on the attribute information and the similarity with the set of words indicating the characteristics of the target person. Since the text recorded in the recording unit for each text category is selected based on the attribute information, a narration text for a person with an appropriate expression taking the situation into consideration can be generated. For example, even in a situation such as a funeral where a positive expression is more desirable than accuracy, an appropriate expression for that situation can be selected. [Brief description of the drawings]

[0007] [Figure 1] FIG. 2 is a diagram showing an example of the functional configuration of a sentence generation device according to the present invention. [Diagram 2] FIG. 4 is a diagram showing an example of a processing flow of the sentence generation device of the present invention. [Diagram 3] A diagram showing examples of sentence categories that make up funeral narration sentences. [Figure 4] FIG. 11 is a diagram showing an example of a processing flow using a tag multi-classification model unit. [Diagram 5] FIG. 2 is a diagram showing an example of the functional configuration of a computer. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] Hereinafter, an embodiment of the present invention will be described in detail. Note that components having the same functions are given the same reference numbers and duplicated explanations will be omitted. EXAMPLES

[0009] FIG. 1 shows an example of the functional configuration of the text generation device of the present invention. FIG. 2 shows an example of the processing flow of the text generation device of the present invention. The text generation device 100 generates a narration text for a person to introduce a person. In particular, it is assumed that a narration text for a funeral is generated. However, the scope of application of the text generation device 100 is not limited to the generation of a text to explain a deceased person at a funeral. As in the case of a funeral, it is intended for a case where it is required to generate a text appropriate to the scene rather than an accurate expression. The text generation device 100 includes at least a recording unit 190, a basic information acquisition unit 110, a word acquisition unit 120, and a text selection unit 130. The text generation device 100 may also include a word replacement unit 140, a funeral parlor information acquisition unit 150, a tag multi-classification model unit 160, a word conversion dictionary 125, and a funeral parlor DB unit 155. In addition, it is sufficient to connect to terminals 810, 820, etc. via a network 900.

[0010] The recording unit 190 records, for each sentence category constituting the person narration sentence, sentences associated with predetermined attribute information and information on tags, which are a set of words corresponding to the sentences, in association with each other. The recorded sentences may be sentences created by a person who can create appropriate sentences according to the situation. FIG. 3 shows an example of sentence categories constituting the funeral narration sentence. In the example shown in FIG. 3, the funeral narration sentence is composed of 16 types of sentence categories. For example, in the pillow word of the first sentence category, the predetermined attribute information may be the date of death and marital experience. The attribute information of the date of death is the month and date. In addition, in the second to sixteenth sentence categories, the predetermined attribute information may be any one of sex, age, marital experience, divorce experience, and whether or not there are children. There may be three types of attribute information for gender: male, female, and common. For age, attribute information can be used that takes into account changes in living conditions, such as 0-3 years old, 4-5 years old, 6-12 years old, 13-18 years old, 19-22 years old, 23-29 years old, ..., 50-59 years old, ..., 100 years old or older. For marital experience, three types of attribute information can be used: married, unmarried, and common. For divorce experience, attribute information can be yes or no. For whether or not the person has children, attribute information can be yes or no.

[0011] The multiple words included in the tag, which is a set of words corresponding to a sentence, are words that express the characteristics of a person. The multiple words included in the tag may include words that are not included in the corresponding sentence. For example, when "She enjoyed her hobby of flower growing, and cherished conversations with her family, and spent her days peacefully" is recorded as a sentence of hobby information in the eighth sentence category, the tag may be set to, for example, "flowers, gardening, family, gentle". This is because when the bereaved family describes the characteristics of the deceased, there is a possibility that a word similar to "gentle" will appear along with the topic of "gardening", and because although the word "gentle" does not appear in this sentence, the gentleness of the deceased is conveyed from the sentence. In addition, as attribute information to be associated with this sentence, gender, age, and marital experience may be set, such as (gender, common), (age, 70 years or older), and (married experience, married). In the recording unit 190, sentences associated with predetermined attribute information and information on tags, which are a set of words corresponding to the sentences, are recorded in association with each other for each sentence category. Therefore, at least several thousand sentences with various expressions must be recorded. The "information about the tag" may be the words themselves contained in the tag, or the result of vectorizing the words. It is sufficient to determine the information so that it is suitable for processing in the sentence generation device 100.

[0012] Further examples of sentences and tags are shown below. For example, possible tags for the sentence "He worked as an office worker until retirement, and after retiring, he devoted himself to farm work as a hobby... I remember the back of (name) who worked hard at everything." are "vegetables, hard work, office worker, field work, farm work." For example, possible tags for the sentence "The memorial photo is a photo of him when he was a little younger, but that was when he was at his most energetic and happy. When I see that smile, when I feel like giving up, I think to myself, "I'll try my best to live!" are "memorial photo, number one, energetic, memories, try my best."

[0013] The basic information acquiring unit 110 acquires predetermined basic information about a person (S110). The basic information includes information that can be used to obtain attribute information that may be used in all text categories. For funerals, the attribute information may include the date of death, sex, age, marital experience, divorce experience, and whether or not the person has children. The basic information may include the name, sex, date of birth, date of death, marital experience, divorce experience, and whether or not the person has children. The "age" included in the attribute information can be obtained from the "date of birth" of the basic information. The text generation device 100 and the terminal 810 may be connected via a network 900, and the basic information acquiring unit 110 may acquire the basic information entered by the bereaved family members into the terminal 810.

[0014] The word acquisition unit 120 acquires a set of words that indicate the characteristics of a target person for the person narration sentence to be generated (S120). The word acquisition unit 120 may acquire words that indicate the characteristics of a person, such as nouns and adjectives, from text data (documents, words, etc.) input to the terminal 810. In the case of a funeral, the word acquisition unit 120 acquires information such as the occupation, hobbies, and personality of the deceased input by the bereaved family to the terminal 810. The sentence generation device 100 may also include a word conversion dictionary 125. The word conversion dictionary 125 records a set of words with similar meanings and a word that represents the set. For example, for a set of words {stylist, barber, hair designer, barber, hairdresser}, the word acquisition unit 120 may record "beautician" as a representative word. If the set of acquired words includes a word recorded by the word conversion dictionary 125, the word acquisition unit 120 converts it to the representative word, and the set of converted words is regarded as the set of acquired words. That is, for example, "hair designer" is converted to "hairdresser." The set may include a representative word. For example, for a set of words {stylist, barber, hair designer, barber, barber, barber}, "hairdresser" may be recorded as a representative word. In this case, "hair designer" is still converted to "hairdresser."

[0015] The sentence generation device 100 sets a sentence category for which sentences are to be generated (S131). The sentence selection unit 130 selects a predetermined number of sentences from the sentences recorded by the recording unit 190 based on the similarity between the set of words included in the tag and the set of words acquired by the word acquisition unit 120 and the attribute information, and sets them as selected sentences (S130). The similarity may be the similarity between a set of sentence word vectors, which are the result of vectorizing each word included in the tag, and a set of acquired word vectors, which are the result of vectorizing each of the multiple words acquired by the word acquisition unit 120. Words may be vectorized using existing technology, and may be converted into vectors of about 300 dimensions, for example. The similarity may be calculated using cosine similarity.

[0016] In the processing flow of the text generation device 100, the text category being generated is also set. In addition, the attribute information of the target person is known from the basic information acquired by the basic information acquisition unit 110. Therefore, the text selection unit 130 selects a predetermined number of texts with high similarity from among the texts recorded in the recording unit 190 as texts of the text category to be generated and whose attribute information associated with the target person is consistent. The "predetermined number" may be appropriately determined. Depending on the text category, selecting one text may be sufficient, and selecting about three texts may be desirable. "Consistent" does not require matching. For example, if the gender of the target person is "female," matching is achieved if the gender of the attribute information associated with the text recorded in the recording unit 190 is "female" or "common."

[0017] It is preferable that the recording unit 190 also records generic sentences that can be selected regardless of similarity for each sentence category. There are cases where a predetermined number of sentences whose similarity meets a predetermined criterion cannot be selected. In such cases, the sentence selection unit 130 selects generic sentences recorded in the recording unit 190 as sentences for the sentence category to be generated. Generic sentences are so-called inoffensive sentences that can be used with anyone without causing any trouble.

[0018] The text generation device 100 may also include a word replacement unit 140. The word replacement unit 140 outputs a text in which words in the selected text are replaced based on basic information (S140). Objective information in the selected text, such as name, date of birth, and date of death, may be replaced by the word replacement unit 140. The text generation device 100 checks whether processing has been performed for the necessary text categories (S135), and ends the processing if completed. The "necessary text categories" refer to text categories used for character narration text among the text categories shown in FIG. 3. For example, for a person who has never been married, the fifth text category, "marriage information," is not necessary. If processing has not been completed, the processing proceeds to the next text category.

[0019] When used for funerals, the text generating device 100 may also include a funeral parlor information acquisition unit 150 and a funeral parlor DB unit 155. The funeral parlor information acquisition unit 150 acquires funeral parlor information that identifies the funeral parlor (S150). The funeral parlor information acquisition unit 150 may acquire the funeral parlor information from information entered by a person related to the funeral to the terminal 820 or the like. The funeral parlor DB unit 155 records the selected text by text category, along with attribute information and tag information, by funeral parlor information. When the funeral parlor DB unit 155 is provided, the text selection unit 130 preferentially selects a predetermined number of texts from the texts recorded in the funeral parlor DB unit with matching funeral parlor information. When the text selection unit 130 cannot select a predetermined number of texts whose similarity meets a predetermined criterion from only the texts recorded in the funeral parlor DB unit with matching funeral parlor information, it also selects texts from the texts recorded in the recording unit 190. By selecting texts by funeral parlor, it is possible to select texts taking into consideration regional characteristics, etc.

[0020] The text generation device 100 may also include a tag multi-classification model unit 160. The tag multi-classification model unit 160 is a trained model that has been trained using pairs of text and tags recorded by the recording unit 190 as teacher data. FIG. 4 shows an example of a processing flow using the tag multi-classification model unit. When creating a new text, it is necessary to create a tag as well. Texts created in the past are associated with tags and recorded in the recording unit 190. If words to be included in tags are determined based on human sensibility alone, individual differences may occur, and even the same person is likely to have differences over time. When a text without a tag is input (S161), the tag multi-classification model unit 160 outputs a tag to be associated with the input text (S160). The output tag may be checked by a person, and if necessary, may be corrected and associated with the text and recorded in the recording unit 190 (S162). After the accuracy of the output of the tag multi-classification model unit 160 has increased, the output tag may be directly associated with the text and recorded in the recording unit 190. Moreover, the tag multi-classification model unit 160 may be provided in a device independent of the text generation device 100. By performing such processing, it is possible to associate objective and consistent tags with text.

[0021] According to the text generation device 100, for each text category constituting a narration text for a person, a text associated with attribute information and information on tags, which are a set of words corresponding to the text, are associated and recorded in the recording unit 190. The recorded text is created by a person who can create a text appropriate for a situation. Then, for each text category, a text is selected based on the similarity with a set of words indicating the characteristics of the target person and the attribute information. Since the text recorded for each text category in the recording unit 190 is selected based on the attribute information, a narration text for a person with an appropriate expression that takes the situation into consideration can be generated. For example, even in a situation such as a funeral where a positive expression is more desirable than accuracy, an appropriate expression for that situation can be selected. Also, compared to a case where a person creates a text, the text can be generated in an extremely short time.

[0022] [Programs, recording media] The various processes described above can be implemented by loading a program that executes each step of the above method into the recording unit 2020 of the computer 2000 shown in FIG. 5, and operating the control unit 2010, input unit 2030, output unit 2040, display unit 2050, etc.

[0023] The program describing the processing contents can be recorded on a computer-readable recording medium, which may be, for example, a magnetic recording device, an optical disk, a magneto-optical recording medium, a semiconductor memory, or any other type of recording medium.

[0024] The program may be distributed, for example, by selling, transferring, lending, etc. portable recording media such as DVDs and CD-ROMs on which the program is recorded. Furthermore, the program may be distributed by storing the program in a storage device of a server computer and transferring the program from the server computer to other computers via a network.

[0025] A computer that executes such a program, for example, first stores the program recorded on a portable recording medium or the program transferred from a server computer in its own storage device. Then, when executing a process, the computer reads the program stored in its own recording medium and executes a process according to the read program. As another execution form of the program, the computer may directly read the program from the portable recording medium and execute a process according to the program, or may execute a process according to the received program each time a program is transferred from the server computer to the computer. The above-mentioned process may be executed by a so-called ASP (Application Service Provider) type service that realizes a processing function only by issuing an execution instruction and obtaining a result, without transferring a program from the server computer to the computer. Note that the program in this embodiment includes information used for processing by an electronic computer and equivalent to a program (data that is not a direct command to the computer but has a nature that specifies the processing of the computer, etc.).

[0026] In addition, in this embodiment, the present device is configured by executing a predetermined program on a computer, but at least a part of the processing contents may be realized by hardware. [Explanation of symbols]

[0027] 100 Text generation device 110 Basic information acquisition unit 120 Word Acquisition Unit 125 Word Conversion Dictionary 130 Sentence selection section 140 Word replacement section 150 Funeral home information acquisition section 155 Funeral home specific DB section 160 Tag multi-classification model section 190 Record section 810,820 Terminals 900 Network

Claims

1. A text generation device for generating a narration text for introducing a person, comprising: a recording unit that records, for each sentence category constituting a narration sentence for a person, sentences associated with predetermined attribute information and information on tags, which are a set of words corresponding to the sentences, in association with each other; a basic information acquiring unit that acquires predetermined basic information regarding the person, including information from which attribute information can be obtained; a word acquisition unit that acquires a set of words that indicate characteristics of a target person of a narration sentence to be generated; a sentence selection unit that selects a predetermined number of sentences for each sentence category from the sentences recorded by the recording unit based on the similarity between a set of words included in the tag and a set of words acquired by the word acquisition unit and whether the attribute information is consistent, and sets the selected sentences as the number of sentences; Equipped with The recording unit also records, for each of the sentence categories, generic sentences that can be selected regardless of similarity, The sentence selection unit selects the generic sentence when the predetermined number of sentences whose similarity satisfies a predetermined criterion cannot be selected for each sentence category. A sentence generation device comprising:

2. A text generation device for generating narration text for introducing a person, comprising: The character narration text is a funeral narration text, a recording unit that records, for each sentence category constituting a narration sentence for a person, sentences associated with predetermined attribute information and information on tags, which are a set of words corresponding to the sentences, in association with each other; a basic information acquiring unit that acquires predetermined basic information regarding the person, including information from which attribute information can be obtained; a word acquisition unit that acquires a set of words that indicate characteristics of a target person of a narration sentence to be generated; a sentence selection unit that selects a predetermined number of sentences for each sentence category from the sentences recorded by the recording unit based on the similarity between a set of words included in the tag and a set of words acquired by the word acquisition unit and whether the attribute information is consistent, and sets the selected sentences as the number of sentences; A funeral parlor information acquisition unit that acquires funeral parlor information for identifying a funeral parlor; a funeral parlor-specific DB section for recording the selected texts by the text category, together with attribute information and tag information, by funeral parlor information; Equipped with The sentence selection unit selects the predetermined number of sentences for each sentence category from sentences recorded in a funeral home DB unit that matches the funeral home information, When the sentence selection unit cannot select the predetermined number of sentences for each sentence category that satisfy a predetermined criterion for similarity from only the sentences recorded in the funeral home DB unit that match the funeral home information, the sentence selection unit also selects sentences from the sentences recorded in the recording unit. A sentence generation device comprising:

3. 3. The text generation device according to claim 1, The plurality of words included in the tag includes words expressing characteristics of a person that are not included in the corresponding sentence. A sentence generation device comprising:

4. A method for generating a narration sentence for introducing a person using a sentence generation device, comprising: A recording unit records in advance, for each sentence category constituting a narration sentence for a person, sentences associated with predetermined attribute information and information on tags, which are a set of words corresponding to the sentences, in association with each other; a basic information acquiring step of acquiring predetermined basic information about the person, including information that allows attribute information of all text categories to be obtained; A word acquisition step of acquiring a set of words indicating characteristics of a target person of a person narration sentence to be generated; a sentence selection step of selecting a predetermined number of sentences for each sentence category from the sentences recorded by the recording unit based on the similarity between a set of words included in the tag and a set of words acquired by the word acquisition unit and whether the attribute information is consistent, and setting the selected sentences as the number of sentences; having The recording unit also records, for each of the sentence categories, generic sentences that can be selected regardless of similarity, The sentence selection step selects the generic sentence when the predetermined number of sentences whose similarity satisfies a predetermined criterion cannot be selected for each sentence category. A sentence generation method comprising:

5. A text generation method for generating a narration text for introducing a person using a text generation device, comprising: The character narration text is a funeral narration text, A recording unit records in advance, for each sentence category constituting a narration sentence for a person, sentences associated with predetermined attribute information and information on tags, which are a set of words corresponding to the sentences, in association with each other; The selected text is recorded in advance in a funeral home DB section for each of the text categories together with attribute information and tag information for each of the funeral home information, A funeral parlor information acquisition step for acquiring funeral parlor information for identifying a funeral parlor; a basic information acquiring step of acquiring predetermined basic information about the person, including information that allows attribute information of all text categories to be obtained; A word acquisition step of acquiring a set of words indicating characteristics of a target person of a person narration sentence to be generated; a sentence selection step of selecting a predetermined number of sentences for each sentence category from the sentences recorded by the recording unit based on the similarity between a set of words included in the tag and a set of words acquired by the word acquisition unit and whether the attribute information is consistent, and setting the selected sentences as the number of sentences; having The sentence selection step includes selecting the predetermined number of sentences for each sentence category from sentences recorded in a funeral home DB unit that matches the funeral home information, In the text selection step, when the predetermined number of texts whose similarity meets the predetermined criterion cannot be selected for each text category from only the texts recorded in the funeral home DB unit that match the funeral home information, texts are also selected from the texts recorded in the recording unit. A sentence generation method comprising:

6. 3. A text generation program for causing a computer to function as the text generation device according to claim 1.

Images