Document creation support device, document creation support method, and program

Abstract

Recently, artificial intelligence has shown high performance in processing tasks such as document classification, translation, summarizing, and Q&A, while exhibiting insufficient performance in creating sentences; it can be said that the need for artificial intelligence lies in assisting people in creating sentences. The purpose of the present invention is to assist in creating a document by assisting people in getting ideas to create new sentences.　A document creation assistance device according to the present invention assists in creating a document, and comprises a morphological analysis unit (12) that subjects document data to morphological analysis to divide the document into words, a masked word setting unit (13) that sets a masked word by masking a specific word among the divided words, and a word retrieval unit (14) that uses a trained natural language processing model (2) to retrieve likelihood considering context and a candidate word on the basis of the masked word, determines a candidate for replacing the masked word according to the likelihood, and completes a candidate document.

Description

[Technical Field]

[0001] The present disclosure relates to a document creation support device, a document creation support method, and a program. [Background technology]

[0002] In the field of human support, artificial intelligence (AI) and machine learning techniques have been attracting attention in recent years. For example, the technology disclosed in Patent Document 1 supports the creation of documents that express problems that people want to solve.

[0003] On the other hand, there are methods (e.g., Word2vec) that use neural networks to represent words as embedding vectors in order to process documents mechanically on a computer. Furthermore, there are language models that take context into account, such as Bidirectional Encoder Representations from Transformers (BERT), which is a general-purpose pre-trained language model that uses large amounts of text data. It has been possible to obtain high performance by fine-tuning it for various natural language tasks. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2020-042695 Summary of the Invention [Problem to be solved by the invention]

[0005] However, while high performance has been achieved in tasks such as document classification, translation, summarization, and Q&A, when it comes to creating text, AI alone is insufficient, and there is a strong need for AI to assist humans.

[0006] The present invention has been made in view of the above points, and has an object to support document creation by supporting human ideas to create new sentences. [Means for solving the problem]

[0007] In order to solve the above problem, the invention of claim 1 is a document creation support device that supports document creation, comprising: a morphological analysis means that performs morphological analysis on data of the document and divides the document into words; a mask word setting means that sets mask words by masking predetermined words among the divided words; and a word search means that uses a trained natural language processing model to search for likelihoods taking into account word candidates and context based on the mask words, and determines replacement candidates for the mask words according to the likelihood, thereby completing candidates for the document. [Effects of the Invention]

[0008] As described above, according to the present invention, the document creation support device 1 can support document creation by determining replacement candidates for words in an input document, thereby supporting human ideas and creating new documents. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a diagram illustrating an electrical hardware configuration of a document creation support device according to an embodiment of the present invention. [Figure 2] 1 is a functional block diagram of a document creation support device according to an embodiment of the present invention; [Figure 3] 10 is a flowchart showing a process for supporting document creation. [Figure 4] FIG. 10 is a diagram illustrating an example of an input document. [Figure 5] FIG. 10 is a diagram showing an example of a word sequence in which a mask word is set. [Figure 6] FIG. 10 is a diagram illustrating an example of document candidates. DETAILED DESCRIPTION OF THE INVENTION

[0010] An embodiment of the present invention will be described below with reference to the accompanying drawings. In this embodiment, a document creation support device will be described that supports document creation by supporting human ideas to create new sentences.

[0011] [Hardware configuration of the document creation support device] Next, the electrical hardware configuration of the document creation support device 1 will be described with reference to Fig. 1. Fig. 1 is a diagram showing the electrical hardware configuration of the document creation support device of this embodiment.

[0012] The document creation support device 1 is a PC (personal computer) and includes, as shown in FIG. 1, a CPU (Central Processing Unit) 101, a ROM (Read Only Memory) 102, and a , RAM (Random Access Memory) 103, SSD (Solid State Drive) 104, Display The computer includes a ray 105 , a keyboard 106 , an external device I / F 107 , a network I / F 108 , a media I / F 109 , and a bus line 110 .

[0013] Of these, the CPU 101 controls the overall operation of the document creation support device 1. The ROM 102 stores programs such as an IPL (Initial Program Loader) used to drive the CPU 101. The RAM 103 is used as a work area for the CPU 101.

[0014] The SSD 104 is a storage device that reads and writes various data such as programs of the document creation support device 1 under the control of the CPU 101. Note that instead of an SSD, a storage device such as an HDD (Hard Disk Drive) may also be used.

[0015] The display 105 is a type of display means such as a liquid crystal display or organic EL (Electro Luminescence) display that displays characters, images, and the like.

[0016] The keyboard 106 is a type of input means having a plurality of keys for inputting characters, numbers, various instructions, and the like.

[0017] The external device I / F 107 is an interface for connecting various external devices, such as an external display as an example of a display unit, a mouse, an external keyboard, or a microphone as an example of an input unit, a printer or a speaker as an example of an output unit, and a USB (Universal Serial Bus) memory as an example of a storage unit.

[0018] The network I / F 108 is a circuit for transmitting and receiving data to and from other evaluation devices via the Internet.

[0019] The media I / F 109 controls reading and writing (storing) of data from and to a recording medium 109m such as a flash memory, etc. The recording medium 109m includes a DVD (Digital Versatile Disc) and a Blu-ray Disc (registered trademark).

[0020] The bus line 110 is an address bus, a data bus, etc. for electrically connecting the components such as the CPU 101 shown in FIG.

[0021] [Functional configuration of the document creation support device] Next, the functional configuration of the document creation support device 1 will be described with reference to Fig. 2. Fig. 2 is a functional block diagram of the document creation support device of this embodiment. As shown in Fig. 2, the document creation support device 1 has a storage unit 10. The storage unit 10 is realized by the RAM 103 or SSD 104 in Fig. 1.

[0022] The document creation support device 1 also includes a document input unit 11, a morphological analysis unit 12, a mask word setting unit 13, a word search unit 14, and a document output unit 15. Each of these units has a function realized by an instruction from the CPU 101 in FIG. 1 based on a program.

[0023] Of these, the document input unit 11 receives input of data of a predetermined document from a user or the like.

[0024] The morphological analysis unit 12 performs morphological analysis on the data of the document input by the document input unit 11 and divides the document into words.

[0025] The mask word setting unit 13 creates a word sequence by masking predetermined words among the words divided by the morphological analysis unit 12 to set mask words.

[0026] The word search unit 14 uses a trained natural language processing model to search for word candidates and likelihoods taking into account the context (before and after the words) based on the mask words (word sequences) set by the mask word setting unit 13, and determines replacement candidates for the mask words according to the likelihoods, thereby completing document candidates. The word search unit 14 also determines replacement candidates with a probability proportional to the likelihood, or selects a predetermined number of words with high likelihoods and randomly determines the replacement candidates from among the selected words. Furthermore, the word search unit 14 determines replacement candidates word by word for multiple mask words. The trained natural language processing model performs machine learning using a known machine learning algorithm based on a neural network, and a model suitable for the field of content of the input document is used here.

[0027] The document output unit 15 sorts and outputs the document candidates completed by the word search unit 14 based on the evaluation index. The types of output include display output displayed on the display 105 in Fig. 1, print output performed on a printer that is an external device via the external device I / F 107 in Fig. 1, and transmission output to another device via the network I / F 108 in Fig. 1.

[0028] [Processing or Operation of the Embodiment] Next, the processing or operation of this embodiment will be described in detail with reference to Figures 3 to 6. Figure 3 is a flowchart showing processing for supporting document creation.

[0029] 3, first, the document input unit 11 inputs document data from the user (S11). Fig. 4 is a diagram showing an example of the input document. Fig. 4 shows a sentence describing the contents of the SDGs (Sustainable Development Goals).

[0030] Next, the morphological analysis unit 12 performs morphological analysis on the document data input by the document input unit 11, and divides the sentence into words (S12). Examples of morphological analysis techniques include MeCab, ChaSen, and KyTea.

[0031] Next, the mask word setting unit 13 selects predetermined words from all the words divided by the morphological analysis unit 12, sets mask words by masking these predetermined words, and creates a word sequence including each mask word (S13). FIG. 5 is a diagram showing an example of a word sequence in which mask words are set. As a method for selecting predetermined words, the mask word setting unit 13 may select words by evaluating the importance of the words using existing methods such as tf-idf or an attention mechanism. tf-idf (Term Frequency-Inverse Document Frequency) is one of the methods for evaluating the importance of words contained in a document, and is mainly used for information processing. It is used in fields such as search and topic analysis. The Attention Mechanism is a mechanism that learns the relationships and points of attention for each element, which is introduced into an Encoder-Decoder model mainly for the purposes of machine translation and image processing. Note that the user may manually select specific words.

[0032] Next, the word search unit 14 inputs the data of the word sequence into a natural language processing model that has been trained in advance using an existing method such as BERT, searches for word candidates and likelihoods that take into account the context (before and after the word), and determines replacement candidates for the masked word probabilistically based on the likelihoods (S14). BERT (Bidirectional Encoder Representations from Transformers) is , a Transformer-based machine learning method for pre-training of natural language processing (NLP). Furthermore, "according to likelihood" may involve determining replacement candidates (words) with a probability proportional to the likelihood, or may involve selecting a predetermined number of words with high likelihood and randomly determining replacement candidates (words) from among the selected words. In this case, the word search unit 14 determines replacement candidates one word at a time for each mask word. The reason for determining replacement candidates one word at a time is that determining one replacement candidate changes the candidates and likelihood of the next mask word. The order in which the word search unit 14 substitutes words may be from the beginning of the sentence or may be random.

[0033] Next, the mask word setting unit 13 determines whether all of the mask words have been replaced (S15). If not all of the mask words have been replaced (S15; NO), the process returns to step S14. On the other hand, if all of the mask words have been replaced (S15; YES), the mask word setting unit 13 completes one document candidate and unmasks (removes) the mask words (S16). Figure 6 is a diagram showing an example of document candidates. In Figure 6, (1) shows the first document candidate, and (2) shows the second document candidate.

[0034] Next, the mask word setting unit 13 determines whether a predetermined number of candidate documents have been created (S17). If the predetermined number of candidate documents have not been created (S17; NO), the process returns to step S13. The predetermined number of candidate documents is preset by the user. On the other hand, if the predetermined number of candidate documents have been created (S17; YES), the document output unit 15 outputs and displays the predetermined number of candidate documents on the display 105 (S18). In this case, the document output unit 15 sorts and displays the documents based on an evaluation index preset by the user. An example of the evaluation index is perplexity, which indicates the predictive performance of the natural language processing model 2. The document output unit 15 may print the documents to a printer, which is an external device, via the external device I / F 107 in FIG. 1, or may transmit the documents to another device via the network I / F 108 in FIG. 1.

[0035] In addition to the documents shown in FIGS. 4 to 6, examples of documents that the document creation support device 1 can support include the following. (1) Using an existing recipe text as input, the system assists in refining recipes by suggesting alternative ingredients instead of words that indicate ingredients, as well as condiments that go well with those ingredients. (2) Supports the creation of new slogans by inputting existing slogans. (3) Uses previously created future forecast scenarios as input to assist in creating further future scenarios.

[0036] As described above, by fine-tuning the trained natural language processing model 2 using sentences from the field, it is expected that more appropriate words will be presented.

[0037] [Major Effects of the Embodiments] As described above, according to this embodiment, the document creation support device 1 uses a trained natural language processing model 2 suitable for the field of the input document to probabilistically search for and present replacement candidates for words that reflect the surrounding context, thereby supporting human ideas and creating new documents.

[0038] 〔supplement〕 The present invention is not limited to the above-described embodiment, and may have the following configurations or processes (operations). (1) The document creation support device 1 can be realized by a computer and a program, but this program can also be recorded on a recording medium or provided via a communication network 100. (2) In the above embodiment, a personal computer is shown as an example of the document creation support device 1, but the document creation support device 1 is not limited to this and may be, for example, a tablet terminal, a smartphone, a smartwatch, or the like. (3) The CPU 101 may be not only a single CPU, but also multiple CPUs. [Explanation of symbols]

[0039] 1. Document creation support device 2. Pre-trained natural language processing model 10 Storage section 11 Document input unit (an example of a document input means) 12 Morphological analysis unit (an example of a morphological analysis means) 13 Mask word setting unit (an example of a mask word setting means) 14 Word search unit (an example of a word search means) 15 Document output unit (an example of a document output means)

Claims

1. A document creation support device that supports document creation, a morphological analysis means for performing a morphological analysis on the data of the document and dividing the document into words; a mask word setting means for setting a mask word by masking a predetermined word among the divided words; a word search means for searching likelihoods of words based on the mask words, taking into account the candidate words and the context, using a trained natural language processing model, and determining replacement candidates for the mask words according to the likelihoods, thereby completing the candidate documents; A document creation support device comprising:

2. the mask word setting means creates a word sequence by setting the mask word; 2. The document creation support device according to claim 1, wherein the word search means uses the trained natural language processing model to search for the likelihood based on the created word sequence.

3. 2. The document creation support device according to claim 1, wherein the word search means determines the replacement candidate with a probability proportional to the likelihood, or selects a predetermined number of words with high likelihood and randomly determines the replacement candidate from among the selected words.

4. 2. The document creation support device according to claim 1, wherein said word search means determines said replacement candidates word by word for a plurality of said mask words.

5. 2. The document creation support device according to claim 1, wherein the trained natural language processing model is BERT.

6. 6. A document creation support device according to claim 1, a document output means for sorting the plurality of document candidates completed by the word search means based on an evaluation index and outputting the sorted document candidates;

7. A document creation support method executed by a document creation support device that supports document creation, comprising: a morphological analysis step of morphologically analyzing the data of the document to divide the document into words; a mask word setting step of setting a mask word by masking a predetermined word among the divided words; a word search step of searching likelihoods of word candidates and contexts based on the mask words using a trained natural language processing model, and determining replacement candidates for the mask words according to the likelihoods, thereby completing the document candidates; A document creation method comprising:

8. A program causing a computer to execute the method according to claim 7.

Images