Summary generation system, summary generation method, and summary generation program

Abstract

To provide a summary generation system, summary generation method, and summary generation program that are capable of generating highly accurate summaries from text data.SOLUTION: A summary generation apparatus 1 includes: an acquisition processing unit 111 configured to acquire text data; a syntactic summary generation processing unit 112 configured to perform syntactic analysis on text of the text data acquired by the acquisition processing unit 111 and generate a first summary including specific words; an AI summary generation processing unit 113 configured to generate a second summary of the text data using a summary generation model generated by machine learning; and an integration processing unit 115 configured to integrate the first and second summaries to generate a summary corresponding to the text data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to techniques for generating summaries from text data. [Background technology]

[0002] Conventionally, systems that generate summaries from text data are known. For example, a system is known that calculates the importance of words included in an original text, inputs the importance to a trained model, and generates a summary of the original text (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-077370 Summary of the Invention [Problem to be solved by the invention]

[0004] However, conventional technologies generate summaries using trained models (AI models) that have been machine-learned using training data (teacher data). This means that the summaries are easily influenced by trends in the training data, and there is a problem in that the accuracy of the generated summaries is low, such as when certain words are misrecognized.

[0005] An object of the present disclosure is to provide a summary generation system, a summary generation method, and a summary generation program that are capable of generating a highly accurate summary from text data. [Means for solving the problem]

[0006] A summary generation system according to one aspect of the present disclosure includes an acquisition processing unit that acquires text data, a first summary generation processing unit that generates a first summary based on the text data acquired by the acquisition processing unit, the first summary including specific words contained in the text data, a second summary generation processing unit that generates a second summary of the text data using a summary generation model generated by machine learning, and an integration processing unit that integrates the first summary and the second summary to generate a summary corresponding to the text data.

[0007] Another aspect of the present disclosure provides a summary generation method in which one or more processors execute the following steps: acquiring text data; generating a first summary based on the text data, the first summary including specific words contained in the text data; generating a second summary of the text data using a summary generation model generated by machine learning; and integrating the first summary and the second summary to generate a summary corresponding to the text data.

[0008] A summary generation program according to another aspect of the present disclosure is a summary generation program for causing one or more processors to execute the following steps: acquiring text data; generating a first summary based on the text data, the first summary including specific words contained in the text data; generating a second summary of the text data using a summary generation model generated by machine learning; and integrating the first summary and the second summary to generate a summary corresponding to the text data. [Effects of the Invention]

[0009] According to the present disclosure, it is possible to provide a summary generation system, a summary generation method, and a summary generation program that are capable of generating a highly accurate summary from text data. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a block diagram showing a configuration of a summary generation system according to an embodiment of the present disclosure. [Figure 2] FIG. 2 is a diagram illustrating an example of text data acquired in the summary generation system according to an embodiment of the present disclosure. [Figure 3] FIG. 3 is a diagram illustrating an example of a syntactic summary generated by the summary generation system according to an embodiment of the present disclosure. [Figure 4] FIG. 4 is a diagram illustrating an example of an AI summary generated in a summary generation system according to an embodiment of the present disclosure. [Figure 5] FIG. 5 is a diagram illustrating an example of a completed summary generated by the summary generation system according to an embodiment of the present disclosure. [Figure 6] FIG. 6 is a diagram showing an example of a determination result page displayed in the summary generation system according to an embodiment of the present disclosure. [Figure 7] FIG. 7 is a flowchart illustrating an example of a procedure of a summary generation process executed in the summary generation system according to an embodiment of the present disclosure. [Figure 8] FIG. 8 is a diagram showing an example of a determination result page displayed in the summary generation system according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. Note that the following embodiments are examples that embody the present disclosure and do not limit the technical scope of the present disclosure.

[0012] The summary generation system according to the present disclosure can be applied to cases where, for example, audio data of a conference is transcribed and converted into text data, and a summary summarizing the contents of the conference is generated from the text data. Note that the text data is not limited to text data obtained by converting audio data into text, but may also be, for example, text data entered into a user terminal (PC) by a conference participant, text data obtained by character recognition of a document image scanned by an image forming device, or text data obtained by translating text in another language.

[0013] FIG. 1 is a block diagram showing the configuration of a summary generation system 10 according to this embodiment. As shown in FIG. 1, the summary generation system 10 includes a summary generation device 1, a user terminal 2, and an audio device 3. The user terminal 2 is an information processing device such as a personal computer or a smartphone, and the audio device 3 is a wireless or wired audio device (microphone speaker device) equipped with a microphone and a speaker. The summary generation device 1 is capable of data communication between the user terminal 2 and the audio device 3 via a network N1. For example, the summary generation device 1 can acquire text data or a document file (e.g., minutes) from the user terminal 2, and can acquire audio data of spoken voice (e.g., conference voice) or text data obtained by converting voice from the audio device 3.

[0014] The summary generation system 10 may be configured by a single summary generation device 1, or may be configured by a combination of the summary generation device 1 and a user terminal 2, or may be configured by a combination of the summary generation device 1 and an audio device 3.

[0015] [Summary Generation Device 1] 1, the summary generation device 1 is an information processing device including a control unit 11, a storage unit 12, an operation display unit 13, and a communication unit 14. The summary generation device 1 may be configured as a personal computer, or may be configured as one or more servers (e.g., cloud servers).

[0016] The communication unit 14 is a communication unit that connects the summary generation device 1 to the network N1 via a wired or wireless connection and performs data communication in accordance with a predetermined communication protocol with external devices such as a user terminal 2 and an audio device 3 via the network N1.

[0017] The operation display unit 13 is a user interface that includes a display unit such as a liquid crystal display or an organic EL display for displaying various information, and an operation unit such as a mouse, keyboard, or touch panel for accepting operations. The operation display unit 13 accepts operations from the administrator of the summary generation device 1.

[0018] The storage unit 12 is a non-volatile storage unit such as an HDD (Hard Disk Drive), an SSD (Solid State Drive), or a flash memory that stores various types of information. The storage unit 12 stores text data input from the user terminal 2, voice data input from the voice device 3, or text data transcribed from voice data. The storage unit 12 also stores summary data generated by the control unit 11.

[0019] The storage unit 12 also stores control programs such as a summary generation program (an example of a summary generation program of the present disclosure) for causing the control unit 11 to execute a summary generation process (see FIG. 7 ) described below. For example, the summary generation program may be non-temporarily recorded on a computer-readable recording medium such as a CD or a DVD, read by a reading device (not shown) such as a CD drive or a DVD drive provided in the summary generation device 1, and stored in the storage unit 12.

[0020] The control unit 11 has control devices such as a CPU, a ROM, and a RAM. The CPU is a processor that executes various types of arithmetic processing. The ROM is a non-volatile storage unit that pre-stores control programs such as a BIOS and an OS that cause the CPU to execute various types of arithmetic processing. The RAM is a volatile or non-volatile storage unit that stores various types of information and is used as a temporary storage memory (work area) for the various types of processing executed by the CPU. The control unit 11 controls the summary generation device 1 by having the CPU execute various control programs pre-stored in the ROM or the storage unit 12.

[0021] Specifically, as shown in FIG. 1, the control unit 11 includes various processing units such as an acquisition processing unit 111, a syntax summary generation processing unit 112, an AI summary generation processing unit 113, a judgment processing unit 114, an integration processing unit 115, and an output processing unit 116. The control unit 11 functions as the various processing units by executing various processes in accordance with the control program using the CPU. Some or all of the processing units may be configured with electronic circuits. The control program may be a program for causing multiple processors to function as the processing units.

[0022] The acquisition processing unit 111 acquires text data. For example, when the audio of a conference is input to the audio device 3, the audio device 3 transcribes the audio, converts it into text, and outputs the text data to the summary generation device 1. The acquisition processing unit 111 acquires the text data from the audio device 3. In another embodiment, the acquisition processing unit 111 acquires text data corresponding to a document input to the user terminal 2.

[0023] 2 shows a specific example of text data A1 corresponding to the voice of a particular conference. The text data A1 includes a plurality of sentences a1 to a17 separated by speakers. Each sentence may consist of a plurality of sentences or a single sentence.

[0024] The syntactic summary generation processing unit 112 generates a summary including specific words contained in the text data based on the text data acquired by the acquisition processing unit 111. Specifically, the syntactic summary generation processing unit 112 syntactically analyzes the text of the text data acquired by the acquisition processing unit 111 to extract specific words, and generates a summary including the extracted specific words (hereinafter referred to as a "syntactic summary"). The syntactic summary is an example of the first summary of the present disclosure. Here, the specific words are words of high importance in the text data and are set according to attributes such as the content and type of the text data. Here, it is assumed that "date" is set as the specific word for text data A1. The specific word may be set by an administrator of the summary generation device 1 or automatically by the control unit 11. A specific example of a method for setting the specific word will be described later.

[0025] The syntactic summary generation processing unit 112 first extracts sentences that include the specific word "date" from sentences a1 to a17 in the text data A1 shown in FIG. 2. Next, the syntactic summary generation processing unit 112 converts each extracted sentence into a predetermined format, for example, a "date: event" format. The event corresponds to the content of the speaker's utterance and is a concise expression of the content of the utterance in one sentence. The syntactic summary generation processing unit 112 may generate the event by omitting particles, conjunctions, etc., or by expressing verbs in their original form.

[0026] The syntactic summary generation processor 112 generates a syntactic summary expressed in the format of "date: event." FIG. 3 shows a specific example of syntactic summary B1 generated by the syntactic summary generation processor 112. Here, syntactic summaries b1 to b11 corresponding to sentences a1 to a17 in the text data A1 (see FIG. 2) are shown. For example, the syntactic summary generation processor 112 generates syntactic summary b1 based on sentence a1, syntactic summary b2 based on sentence a2, and syntactic summary b6 based on sentence a8. In addition, the syntactic summary generation processor 112 excludes sentences that do not include a date (e.g., sentences a7, a12, a14, a15, and a16) from the syntactic summary.

[0027] Furthermore, the syntactic summary generation processor 112 generates a syntactic summary for a sentence that includes at least one specific word. For example, if a sentence includes only one date word, the syntactic summary generation processor 112 extracts the date and generates a syntactic summary for the sentence (e.g., syntactic summaries b1, b6, b7, b9 to b11), and if a sentence includes multiple date words, the syntactic summary generation processor 112 extracts multiple dates and generates a syntactic summary for the sentence (e.g., syntactic summaries b2 to b5, b8).

[0028] In this way, the syntactic summary generation processor 112 extracts sentences that include a specific word (here, "date") from the sentences a1 to a17 and generates syntactic summary B1 (b1 to b11) represented in a predetermined format. The syntactic summary generation processor 112 is an example of a first summary generation processor of the present disclosure.

[0029] The AI ​​summary generation processing unit 113 generates a summary of text data (hereinafter referred to as an "AI summary") using a summary generation model (AI model) generated by machine learning. The AI ​​summary is an example of a second summary of the present disclosure. For example, the summary generation model is generated by machine learning learning data of various text data. The summary generation model may be generated in the summary generation device 1, or may be generated by an external device (learning device) and downloaded to the summary generation device 1. Furthermore, the summary generation device 1 may access a cloud server such as the learning device to use the summary generation model.

[0030] Machine learning includes algorithms such as supervised learning using supervised data, unsupervised learning using unsupervised data, and reinforcement learning. Furthermore, to realize these methods, a method called "deep learning" is used, which learns to extract features themselves. In this embodiment, trained models based on the various algorithms described above are provided. For example, machine learning is performed using the supervised data and unsupervised data as input data (learning data), and a summary generation model that executes a summary generation process is generated. In this embodiment, the AI ​​summary generation processing unit 113 may use a well-known summary generation model.

[0031] Figure 4 shows a specific example of AI summary C1 generated by the AI ​​summary generation processing unit 113. Here, AI summaries c1 to c4 corresponding to sentences a1 to a17 in text data A1 (see Figure 2) are shown. The AI ​​summary generation processing unit 113 inputs sentences a1 to a17 into a summary generation model and obtains AI summaries c1 to c4 generated by the summary generation model. AI summaries c1 to c4 are new sentences created based on sentences a1 to a17.

[0032] In this way, the AI ​​summary generation processing unit 113 generates one or more AI summaries using all the sentences in the text data A1. Here, the AI ​​summary generation processing unit 113 may generate an AI summary of a sentence that includes a specific word and an AI summary of a sentence that does not include the specific word. In the example shown in FIG. 4, the AI ​​summary generation processing unit 113 generates AI summaries c2 to c4 that include a date and an AI summary c1 that does not include a date. The AI ​​summary generation processing unit 113 is an example of a second summary generation processing unit of the present disclosure.

[0033] Here, AI models (trained models) such as summary generation models are easily affected by trends in the training data, and there is a problem of misrecognizing specific words. For example, when text data including a date is input into a summary generation model to generate a summary, the date may be misrecognized. Misrecognition of important words contained in the text data will undermine the reliability of the summary. In response to this, this embodiment includes a process for determining whether a date is incorrect in the AI ​​summary generated by the AI ​​summary generation processing unit 113.

[0034] Specifically, when the AI ​​summary contains a specific word (here, a date), the determination processing unit 114 refers to the text data and determines whether the date is correct. Specifically, the determination processing unit 114 extracts the date by syntactically analyzing the AI ​​summary C1 (see FIG. 4) generated by the AI ​​summary generation processing unit 113. For example, the determination processing unit 114 extracts the date by syntactically analyzing each of the AI ​​summaries c1 to c4. Here, the determination processing unit 114 extracts "April 15th" from the AI ​​summary c2, "May 15th" from the AI ​​summary c3, and "April 10th" from the AI ​​summary c4. Because the AI ​​summary c1 does not contain a date, the determination processing unit 114 excludes the AI ​​summary c1 from the determination target.

[0035] Next, the determination processing unit 114 determines whether the date "April 15th" extracted from the AI ​​summary c2 is included in the text data A1 (see FIG. 2). Here, the determination processing unit 114 determines that the date "April 15th" is not included in any of the sentences a1 to a17 of the text data A1.

[0036] Similarly, the determination processing unit 114 determines whether the date "May 15th" extracted from AI summary c3 is included in text data A1. Here, the determination processing unit 114 determines that the date "May 15th" is included in sentence a6 of text data A1. The determination processing unit 114 also determines that the date "April 10th" extracted from AI summary c4 is included in sentence a9 of text data A1. Therefore, the determination processing unit 114 determines that "April 15th" is incorrect, and determines that "May 15th" and "April 10th" are correct.

[0037] Furthermore, if the text data contains a word that matches the specific word, the determination processing unit 114 determines that the AI ​​summary containing the specific word is correct; if the text data does not contain a word that matches the specific word, the determination processing unit 114 determines that the AI ​​summary containing the specific word is incorrect. For example, the date "May 15th" extracted from AI summary c3 matches the date included in sentence a6 of text data A1, so the determination processing unit 114 determines that AI summary c3 is correct. Furthermore, the date "April 10th" extracted from AI summary c4 matches the date included in sentence a9 of text data A1, so the determination processing unit 114 determines that AI summary c4 is correct. In contrast, the date "April 15th" extracted from AI summary c2 does not match any date included in text data A1, so the determination processing unit 114 determines that AI summary c2 is incorrect.

[0038] The integration processing unit 115 generates a summary (completed summary) corresponding to the text data by integrating the syntactic summary generated by the syntactic summary generation processing unit 112 and the AI ​​summary generated by the AI ​​summary generation processing unit 113. Specifically, the integration processing unit 115 integrates the syntactic summary B1 (see FIG. 3) generated by the syntactic summary generation processing unit 112 based on the text data A1 (see FIG. 2) with the AI ​​summary C1 (see FIG. 4) generated by the AI ​​summary generation processing unit 113 based on the text data A1 (see FIG. 2) to generate a completed summary D1 (see FIG. 5) corresponding to the text data A1.

[0039] Here, when the determination processing unit 114 determines that a specific word (date) is incorrect, the integration processing unit 115 generates a completed summary by excluding the corresponding AI summary. For example, in the above example, the determination processing unit 114 determines that AI summary c2 is incorrect, so the integration processing unit 115 excludes AI summary c2 from AI summaries c1 to c4 and generates a completed summary D1 (see Figure 5) by combining the remaining AI summaries c1, c3, and c4 with each of the syntax summaries b1 to b11.

[0040] In this way, when the AI ​​summary generation processing unit 113 generates an AI summary of one sentence containing a specific word (date) (e.g., AI summaries c3 and c4) and an AI summary of one sentence not containing the specific word (e.g., AI summary c1), the integration processing unit 115 generates a completed summary by integrating the AI ​​summary containing the correct specific word, the AI ​​summary not containing the specific word, and each syntactic summary (e.g., syntactic summaries b1 to b11).

[0041] This allows us to exclude low-confidence sentences (AI summary c2) and generate an appropriate summary consisting only of highly reliable sentences.

[0042] The output processing unit 116 outputs the completed summary generated by the integration processing unit 115. For example, the output processing unit 116 displays the completed summary D1 (see FIG. 5 ) on the operation display unit 13. Furthermore, for example, when the organizer of a conference requests the generation of a summary on his / her own user terminal 2, the output processing unit 116 may transmit data of the completed summary D1 to the user terminal 2, or may display a web page of the completed summary D1 on the user terminal 2.

[0043] Furthermore, when the determination processing unit 114 determines that a specific word is incorrect, the output processing unit 116 may display the AI ​​summary to be excluded and the specific word in a distinguishable manner. For example, as shown in FIG. 6, the output processing unit 116 displays the AI ​​summary c2 determined to be incorrect by the determination processing unit 114 in the determination result page P1 so that the user can distinguish it, for example, by underlining it. Furthermore, the output processing unit 116 displays the date determined to be incorrect by the determination processing unit 114 in the determination result page P1 so that the user can distinguish it, for example, by surrounding it with a frame. This allows the user to easily recognize the incorrect summary and date among the AI ​​summaries.

[0044] [Summary generation process] FIG. 7 shows an example of the procedure of the summary generation process executed by the control unit 11 of the summary generation device 1.

[0045] The present disclosure can be understood as a summary generation method (the summary generation method of the present disclosure) that executes one or more steps included in the summary generation process. Furthermore, one or more steps included in the summary generation process described herein may be omitted as appropriate. Furthermore, the steps in the summary generation process may be executed in a different order as long as the same operational effect is achieved. Furthermore, while the description here takes as an example a case in which the control unit 11 executes each step in the summary generation process, in other embodiments, one or more processors may execute each step in the summary generation process in a distributed manner.

[0046] <Step S1> First, in step S1, the control unit 11 acquires text data for which a summary is to be generated. For example, the control unit 11 acquires text data A1 (see FIG. 2) obtained by converting the speech of a conference into text, which is input from an external device. In another embodiment, if the control unit 11 has speech recognition and character conversion functions, the control unit 11 may acquire speech data input from an external device and convert it into text data.

[0047] When the control unit 11 acquires text data, it executes in parallel two processes: a "syntax summarization process S10 (steps S11 to S13)" that analyzes the text data and generates a first summary (syntax summary) containing a specific word (here, "date"), and an "AI summarization process S20 (steps S21 to S26)" that generates a second summary (AI summary) of the text data using a summary generation model generated by machine learning. Note that the order of the syntactic summarization process S10 and the AI ​​summarization process S20 does not matter.

[0048] <Step S11> In the syntactic summarization process S10, in step S11, the control unit 11 performs syntactic analysis on the text of the text data. For example, for each of sentences a1 to a17 in the text data A1 shown in Fig. 2, the control unit 11 analyzes the structure of the sentence, such as words, phrases, symbols, numbers (dates, times, etc.), subjects, predicates, modifiers, nouns, particles, and verbs contained in the sentence.

[0049] <Step S12> In step S12, the control unit 11 extracts dates from the results of the syntax analysis. For example, the control unit 11 extracts "March 15th" from sentence a1 of the text data A1 (see FIG. 2), "March 1st" and "March 31st" from sentence a2, and "mid-March" and "April 5th" from sentence a3.

[0050] <Step S13> In step S13, the control unit 11 generates a syntactic summary. Specifically, the control unit 11 extracts all sentences containing dates from the text data A1, and converts each extracted sentence into a predetermined format ("date: event") using the date and the results of the syntactic analysis to generate a syntactic summary. For example, when the control unit 11 extracts a sentence a1 containing "March 15th" from the text data A1, it generates a syntactic summary b1 (see FIG. 3) of "March 15th: Reporting meeting starts on March 15th." For example, when the control unit 11 extracts a sentence a2 containing "March 1st" and "March 31st" from the text data A1, it generates a syntactic summary b2 (see FIG. 3) of "March 1st, March 31st: Loan performance from March 1st to March 31st, results exceeding the target by 5%." After step S13, the control unit 11 transitions the process to step S3.

[0051] <Step S21> In the AI ​​summarization process S20, in step S21, the control unit 11 generates an AI summary based on the text data. For example, the control unit 11 inputs text data A1 (see FIG. 2) into a summary generation model (AI model) generated by machine learning to generate an AI summary. The control unit 11 acquires the AI ​​summary C1 (see FIG. 4) generated by the summary generation model.

[0052] <Step S22> In step S22, the control unit 11 parses the AI ​​summary C1 generated in step S21. For example, the control unit 11 parses each of the AI ​​summaries c1 to c4 (see FIG. 4).

[0053] <Step S23> In step S23, the control unit 11 extracts dates from the results of the parsing performed in step S22. For example, the control unit 11 extracts dates from the results of parsing AI summaries c1 to c4. Here, the control unit 11 extracts "April 15th" from AI summary c2, "May 15th" from AI summary c3, and "April 10th" from AI summary c4.

[0054] <Step S24> In step S24, the control unit 11 determines whether the date extracted in step S23 is included in the text data A1. If the control unit 11 determines that the date is included in the text data A1 (S24: Yes), the control unit 11 proceeds to step S25. On the other hand, if the control unit 11 determines that the date is not included in the text data A1 (S24: No), the control unit 11 proceeds to step S26.

[0055] In another embodiment, the control unit 11 may determine whether or not the date is included in the syntactic summary B1 (see FIG. 3) generated in the syntactic summarization process S10.

[0056] <Step S25> In step S25, the control unit 11 adopts the AI ​​summary as the summary to be finally output. For example, the date "May 15th" extracted from AI summary c3 is included in sentence a6 of text data A1, and the date "April 10th" extracted from AI summary c4 is included in sentence a9 of text data A1, so the control unit 11 adopts AI summary c3 and AI summary c4. Note that the control unit 11 also adopts an AI summary that does not include a date (for example, AI summary c1 shown in FIG. 4) as a summary. After step S25, the control unit 11 transitions the process to step S3.

[0057] <Step S26> In step S26, the control unit 11 excludes the AI ​​summary from the summary to be finally output. For example, since the date "April 15th" extracted from AI summary c2 is not included in any of sentences a1 to a17 of text data A1, the control unit 11 determines that the date "April 15th" and AI summary c2 are incorrect and excludes (deletes) AI summary c2. After step S26, the control unit 11 transitions the process to step S3.

[0058] <Step S3> In step S3, the control unit 11 integrates the syntactic summary generated in the syntactic summarization process S10 and the AI ​​summary generated in the AI ​​summarization process S20. Specifically, the control unit 11 integrates the syntactic summary B1 (see FIG. 3) generated based on the text data A1 (see FIG. 2) with the AI ​​summary C1 (see FIG. 4) generated based on the text data A1 (see FIG. 2) to generate a completed summary D1 (see FIG. 5) corresponding to the text data A1. Here, the control unit 11 integrates the AI ​​summaries c1, c3, and c4, excluding the AI ​​summary c2 with an incorrect date, with each of the syntactic summaries b1 to b11, to generate the completed summary D1 (see FIG. 5).

[0059] <Step S4> In step S4, the control unit 11 outputs the integrated summary (completed summary). For example, the control unit 11 displays the completed summary D1 (see FIG. 5) on the operation display unit 13. Furthermore, as shown in FIG. 6, the control unit 11 may display the AI ​​summary c2 determined to be incorrect and the date "April 15th" in a distinguishable manner on the determination result page P1.

[0060] The control unit 11 executes the summary generation process in the above manner. The control unit 11 also repeatedly executes the above process every time text data is acquired.

[0061] As described above, the summary generation system 10 according to the present disclosure acquires text data and performs syntactic analysis of the text of the text data to generate a syntactic summary (first summary) that includes a specific word. Furthermore, the summary generation system 10 generates an AI summary (second summary) of the text data using a summary generation model generated by machine learning. The summary generation system 10 then integrates the syntactic summary and the AI ​​summary to generate a summary (completed summary) corresponding to the text data. Furthermore, when a specific word is included in the syntactic summary, the summary generation system 10 refers to the text data to determine whether the specific word is correct, and if it is determined that the specific word is incorrect, generates a complete summary by excluding the AI ​​summary.

[0062] According to the above configuration, accurate summaries can be obtained by syntactic analysis. Furthermore, it is possible to determine whether the summaries generated by the AI ​​model are correct, and obtain only the correct summaries. Then, the accurate summaries obtained by syntactic analysis and the error-free summaries generated by the AI ​​model can be integrated to generate a final summary. This makes it possible to generate highly accurate summaries from text data.

[0063] [Other embodiments] In another embodiment of the present disclosure, the control unit 11 may display, on the determination result page P1, the location of the AI ​​summary c2 determined to be incorrect (e.g., "the second line"), the incorrect date "April 15th," and the candidate correct date "April 5th," as shown in FIG. 8. The control unit 11 may also display a correction button K1 that accepts a correction operation for the AI ​​summary c2 determined to be incorrect. For example, the user can press the correction button K1 to correct "April 15th" to "April 5th." The control unit 11 may adopt the corrected AI summary c2 as the completed summary.

[0064] In the above-described embodiment, the control unit 11 excludes the AI ​​summary (AI summary c2 in the above example) that includes the date determined to be incorrect from the completed summary. However, in another embodiment, the control unit 11 may exclude the syntactic summary (syntactic summary b3 in FIG. 3) that corresponds to the AI ​​summary (AI summary c2) that includes the date determined to be incorrect from the completed summary. The control unit 11 may also exclude both AI summary c2 and syntactic summary b3. The control unit 11 may also exclude one of AI summary c2 and syntactic summary b3, selected by the user.

[0065] In another embodiment, the control unit 11 may correct the date determined to be incorrect to a correct date based on the text data. For example, the control unit 11 may identify, from among multiple syntactic summaries, a syntactic summary that corresponds to (is similar to) the AI ​​summary that includes the date determined to be incorrect, and correct the date determined to be incorrect to the date included in the identified syntactic summary. In this case, the control unit 11 may adopt the corrected AI summary as the completed summary.

[0066] In the above-described embodiment, "date" is used as an example of a specific word, but the specific word of the present disclosure is not limited to this. The control unit 11 may also be configured to extract specific words without including a syntactic analysis process. For example, rule-based extraction may be used, extracting specific words according to specific rules, such as extracting words containing numbers and "month" and "day." Furthermore, if there are important keywords other than dates, such as product names, candidates for these keywords may be stored, and specific words may be extracted by matching them on a string basis. In these cases, the control unit 11 may include an extracted word summary generation processing unit instead of the syntactic summary generation processing unit 112. The syntactic summary generation processing unit 112 and the extracted word summary generation processing unit are examples of a first summary generation processing unit of the present disclosure.

[0067] Furthermore, the control unit 11 may set specific words based on the attributes of the text data. For example, if the text data is data related to a contract, the control unit 11 may set specific words such as "articles" of the contract-related law, "company name," and "contract period." In this way, the control unit 11 may set specific words based on the type of document corresponding to the text data, the subject of the conversation, the topic, etc. Furthermore, the control unit 11 may set specific words such as "number of people," "percentage (%)," "amount," and "product name."

[0068] In the above embodiment, the control unit 11 corrects the correction portion, but in another embodiment, the control unit 11 may present the correction content as a list instead of directly changing the correction content. In this case, the control unit 11 may correct the relevant portion in response to a selection operation by the user.

[0069] [Disclosure Note] The following is a summary of the disclosure extracted from the above-described embodiment. Note that the configurations and processing functions described in the following supplementary notes can be selected and combined as desired.

[0070] <Appendix 1> an acquisition processing unit that acquires text data; a first summary generation processing unit that generates a first summary based on the text data acquired by the acquisition processing unit, the first summary including specific words included in the text data; a second summary generation processor that generates a second summary of the text data using a summary generation model generated by machine learning; an integration processing unit that integrates the first summary and the second summary to generate a summary corresponding to the text data; A summary generation system comprising:

[0071] <Appendix 2> the first summary generation processing unit parses the text of the text data to extract the specific words, and generates the first summary including the specific words. 10. The summary generation system of claim 1.

[0072] <Appendix 3> a determination processing unit that, when the specific word is included in the second summary, determines whether the specific word is correct by referring to the text data; the integration processing unit generates the summary while excluding the second summary when the specific word is determined to be incorrect. 3. The summary generation system according to claim 1 or 2.

[0073] <Appendix 4> the determination processor determines that the second summary is correct if the text data includes a word that matches the specific word, and determines that the second summary is incorrect if the text data does not include a word that matches the specific word. 3. A summary generation system as described in Appendix 3.

[0074] <Appendix 5> the first summary generation processing unit generates a sentence including at least one of the specific words as the first summary; 5. A summary generation system according to any one of appendices 1 to 4.

[0075] <Appendix 6> When the second summary generation processing unit generates the second summary of a sentence including the specific word and the second summary of a sentence not including the specific word, the integration processing unit generates the summary by integrating the second summary including the correct specific word, the second summary not including the specific word, and the first summary. 3. A summary generation system as described in Appendix 3.

[0076] <Appendix 7> an output processing unit that outputs the summary generated by the integration processing unit; the output processing unit, when the determination processing unit determines that the specific word is incorrect, displays the second summary to be excluded and the specific word determined to be incorrect in a distinguishable manner. 10. The summary generation system of claim 3, 4 or 6.

[0077] <Appendix 8> The specific word is a word representing a date. 8. A summary generation system according to any one of appendices 1 to 7.

[0078] <Appendix 9> setting the specific word based on the attribute of the text data; A summary generation system according to any one of appendices 1 to 8. [Explanation of symbols]

[0079] 1: Summary generator 2: User device 3: Audio equipment 10: Summary generation system 11: Control section 12: Storage section 13: Operation display section 14: Communications Department 111: Acquisition processing unit 112: Syntax summary generation processing unit 113: AI summary generation processing unit 114: Judgment processing unit 115: Integrated processing unit 116: Output processing section

Claims

1. an acquisition processing unit that acquires text data; a first summary generation processing unit that generates a first summary based on the text data acquired by the acquisition processing unit, the first summary including specific words included in the text data; a second summary generation processor that generates a second summary of the text data using a summary generation model generated by machine learning; an integration processing unit that integrates the first summary and the second summary to generate a summary corresponding to the text data; A summary generation system comprising:

2. the first summary generation processing unit syntactically analyzes the text of the text data to extract the specific word, and generates the first summary including the specific word. The summary generation system of claim 1 .

3. a determination processing unit that, when the specific word is included in the second summary, determines whether the specific word is correct by referring to the text data; the integration processing unit generates the summary while excluding the second summary when the specific word is determined to be incorrect. The summary generation system of claim 1 .

4. the determination processing unit determines that the second summary is correct when the text data includes a word that matches the specific word, and determines that the second summary is incorrect when the text data does not include a word that matches the specific word. The summary generation system according to claim 3 .

5. the first summary generation processing unit generates a sentence including at least one of the specific words as the first summary; The summary generation system of claim 1 .

6. When the second summary generation processing unit generates the second summary of a sentence including the specific word and the second summary of a sentence not including the specific word, the integration processing unit generates the summary by integrating the second summary including the correct specific word, the second summary not including the specific word, and the first summary. The summary generation system according to claim 3 .

7. an output processing unit that outputs the summary generated by the integration processing unit; the output processing unit, when the determination processing unit determines that the specific word is incorrect, displays the second summary to be excluded and the specific word determined to be incorrect in a distinguishable manner. The summary generation system according to claim 3 .

8. The specific word is a word representing a date. The summary generation system of claim 1 .

9. setting the specific word based on the attribute of the text data; The summary generation system according to any one of claims 1 to 8.

10. Acquiring text data; generating a first summary based on the text data, the first summary including specific words contained in the text data; generating a second summary of the text data using a summary generation model generated by machine learning; combining the first summary and the second summary to generate a summary corresponding to the text data; The summary generation method is executed by one or more processors.

11. Acquiring text data; generating a first summary based on the text data, the first summary including specific words contained in the text data; generating a second summary of the text data using a summary generation model generated by machine learning; combining the first summary and the second summary to generate a summary corresponding to the text data; a summary generating program for causing one or more processors to execute the above;

Images