Information processing program, information processing method, and information processing device

Abstract

The invention relates to an information processing program, an information processing method, and an information processing apparatus. The information processing apparatus calculates vectors of a plurality of continuous sentences and of a plurality of sentences having a relationship among preceding and succeeding sentences, respectively. The information processing device generates a machine learning model by sequentially inputting vectors of a plurality of sentences into the machine learning model and training the vectors, and when a vector of a certain sentence is input into the machine learning model, the machine learning model predicts a sentence vector of a sentence input next to the certain sentence. The information processing apparatus calculates a vector of a first sentence and a vector of a second sentence following the first sentence. The information processing apparatus calculates a vector of a sentence predicted to follow a first sentence by inputting the vector of the first sentence into a machine learning model, and determines whether a vector of a second sentence is appropriate.

Description

Technical Field

[0001] The present invention relates to an information processing program and the like. Background Art

[0002] In recent years, a service has been provided that calculates the vector of a sentence and uses the calculated vector to perform various processes such as translation into other languages ​​and database search. However, if there is an input error in the sentence specified by the user, the vector of the sentence cannot be calculated with high accuracy, and errors may occur in the translation, search, and other processes.

[0003] For example, a proper sentence "its function is a feature" and an incorrectly input sentence "its yesterday is a feature" are sentences with greatly different meanings, and the vectors of the sentences are also greatly different.

[0004] As a previous technology for correcting input errors in sentences, there is the following previous technology, which uses a data set of pairs of input errors and the corrected sentences from the correction history to pre-train a learning model, and inputs the sentence as the object into the trained learning model, thereby correcting the input error of the sentence as the object.

[0005] Patent Document 1: Japanese Patent Application Publication No. 2019-101993

[0006] Non-patent document 1: Kazuhiro Miki and others, "A solution to the English blank completion problem using BERT", Department of Information Science, Faculty of Engineering, Okayama University, DEIM2020G2-4 (day1p47)

[0007] Non-Patent Literature 2: Tanaka Yu and others, "Construction of a Japanese Input Error Dataset Using Wikipedia's Correction History," Kyoto University Graduate School of Information Science, Language Processing Society, 26th Annual Conference, Proceedings, March 2020

[0008] However, in the above-mentioned prior art, it is a technology for filling in blank spaces such as a part of words in an article, and although the accuracy of words filled in blank spaces of a sentence composed of multiple words is high, there are few high-precision descriptions related to sentences filled in blank spaces of an article composed of multiple sentences, and sentences containing input errors are not detected. In addition, in the prior art, although input errors such as wrong characters and missing characters can be corrected, there are many cases where input errors that are incorrectly converted cannot be correctly corrected. Summary of the invention

[0009] In one aspect, the present invention aims to provide an information processing program, an information processing method, and an information processing device capable of estimating sentences to fill in blanks in a text composed of a plurality of sentences and detecting sentences containing input errors.

[0010] In the first technical solution, a computer is made to perform the following processing. The computer calculates vectors of a plurality of consecutive sentences that are related to each other in the preceding and following sentences. The computer generates a machine learning model by inputting the vectors of the plurality of sentences into the machine learning model in sequence and training the model, wherein when the vector of a certain sentence is input into the machine learning model, the machine learning model predicts the sentence vector of the sentence input next to the certain sentence. The computer calculates the vector of the first sentence and the vector of the second sentence following the first sentence. The computer calculates the vector of the sentence predicted to follow the first sentence by inputting the vector of the first sentence into the above-mentioned machine learning model, and determines whether the vector of the second sentence is appropriate.

[0011] It is possible to estimate sentences that fill in blanks in a text composed of multiple sentences and detect sentences containing input errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a diagram for explaining the processing in the learning phase of the information processing device involved in this embodiment.

[0013] Figure 2 This is a diagram for explaining the processing in the analysis phase of the information processing device involved in this embodiment.

[0014] Figure 3 It is a functional block diagram showing the structure of the information processing device involved in this embodiment.

[0015] Figure 4 This is a diagram showing an example of the data structure of a word vector dictionary.

[0016] Figure 5A FIG. (1) is a diagram for explaining the process of calculating a sentence vector.

[0017] Figure 5B FIG. (2) is a diagram for explaining the process of calculating a sentence vector.

[0018] Figure 6 This is a diagram for explaining the process of generating a sentence inverted index.

[0019] Figure 7 This is a flowchart showing the processing procedure of the learning phase of the information processing device involved in this embodiment.

[0020] Figure 8 This is a flowchart showing the processing procedure of the analysis phase of the information processing device involved in this embodiment.

[0021] Fig. 9 FIG. (1) is a diagram for explaining other processing of the information processing device.

[0022] Fig.10 FIG. (2) is a diagram for explaining other processing of the information processing device.

[0023] Fig.11 This is a diagram showing an example of the hardware configuration of a computer that realizes the same function as the information processing device of the embodiment. DETAILED DESCRIPTION

[0024] Hereinafter, embodiments of the information processing program, information processing method, and information processing device disclosed in the present application will be described in detail based on the drawings. However, the present invention is not limited to the embodiments.

[0025] Example

[0026] The processing of the information processing device according to this embodiment will be described. After executing the processing of the learning phase, the information processing device executes the processing of the analysis phase. Figure 1 This is a diagram for explaining the processing in the learning phase of the information processing device involved in this embodiment.

[0027] In the learning phase, the information processing device performs learning of the machine learning model 50 (training the machine learning model 50) using a plurality of articles included in the teacher data 141. The machine learning model 50 is a NN (Neural Network) such as BERT (Pre-training of Deep Bidirectional Transformers for Language Understanding), Next Sentence Prediction, and Transformers.

[0028] The text included in the teacher data 141 includes a plurality of sentences. The plurality of sentences have a predetermined relationship between the preceding and following sentences. Each sentence is a sentence set in advance based on the syllogism method of induction, deduction, etc.

[0029] For example, the article 10a contains the sentence "Birds are born with eggs.", "Birds are born with eggs.", ..., and "Birds are born with eggs." are included in this order from the beginning. Article 10b contains the sentence "Bird of the bird's egg is born," in order from the beginning.

[0030] The information processing device calculates the sentence vector of each sentence included in the sentences 10a, 10b and other sentences. For example, the information processing device performs morpheme analysis on the sentence to divide it into words, and accumulates the vectors of each word to calculate the sentence vector.

[0031] Set the sentence vector of the sentence "A bird lays eggs." in article 10a as "SV1-1". Set the sentence vector of the sentence "A penguin is a bird." as "SV1-2". Set the sentence vector of the sentence "Therefore, a penguin lays eggs." as "SV1-3".

[0032] Set the sentence vector of the sentence "A bird is born from an egg." in article 10b as "SV2-1". Set the sentence vector of the sentence "A pigeon is a kind of bird." as "SV2-2". Set the sentence vector of the sentence "Therefore, a pigeon is born from an egg." as "SV2-3".

[0033] The information processing device repeatedly executes the process of sequentially inputting the vectors of the sentences at the beginning included in the article into the machine learning model 50. For example, the information processing device inputs the sentence vectors into the machine learning model 50 in the order of the sentence vectors "SV1-1", "SV1-2",..., "SV1-3". The information processing device inputs the sentence vectors into the machine learning model 50 in the order of the sentence vectors "SV2-1", "SV2-2",..., "SV2-3".

[0034] The information processing device generates the machine learning model 50 by executing the process in the above learning stage. When the sentence vector of a certain first sentence is input, this machine learning model predicts the sentence vector of the next second sentence of the first sentence.

[0035] Figure 2 This is a diagram for explaining the process in the analysis stage of the information processing device according to the present embodiment. In the analysis stage, the information processing device uses the trained machine learning model 50 to calculate the sentence vectors included in the article to be processed, and detects inappropriate sentences by means of cosine similarity, etc.

[0036] In Figure 2 the description, the article to be processed including input errors, etc. is set as article 20. In article 20, it is composed of the sentence "A bird lays eggs.", the sentence "A penguin takes pictures.",..., the sentence "Therefore, a penguin lays eggs." in sequence from the beginning. The sentence "A penguin takes pictures." is a sentence with an input error including the homophone "takes pictures" of the word "bird" compared to the correct sentence "A penguin is a bird." included in the teacher data 141 of article 10a.

[0037] The information processing device calculates the sentence vector "SV1-1" of the sentence "A bird lays eggs.", and inputs the calculated sentence vector "SV1-1" into the machine learning model 50, thereby predicting the sentence vector of the next sentence of the sentence "A bird lays eggs.". In Figure 2In the example shown, the sentence vector of "SV1-2" is predicted as the next sentence of the sentence "鳥は卵を産む。" by the machine learning model 50.

[0038] The information processing device calculates the sentences included in the article 20 and is the sentence vector "SV3" of the next sentence "ペンギンは撮り" of the sentence "鳥は卵を産む。".

[0039] The information processing device calculates the cosine similarity between the sentence vector "SV1-2" of the next sentence predicted by the machine learning model 50 and the sentence vector "SV3" of the sentence "ペンギンは撮り", which is the sentence included in the article 20 and is the next sentence of the sentence "鳥は卵を産む。".

[0040] When the cosine similarity is less than the threshold value, the information processing device determines that the sentence "ペンギンは鳥。", which is the sentence included in the article 10a and is the next sentence of the sentence "鳥は卵を産む。", is a correct (hereinafter referred to as appropriate) sentence. On the other hand, when the cosine similarity is less than the threshold value, the information processing device determines that the sentence "ペンギンは撮り。", which is the sentence included in the article 20 and is the next sentence of the sentence "鳥は卵を産む。", is an inappropriate sentence including input errors or the like.

[0041] As described above, the information processing device generates the machine learning model 50 by sequentially inputting the vectors of the sentences of the article included in the teacher data 141 into the machine learning model 50. When the sentence vector of a certain first sentence is input, the machine learning model predicts the sentence vector of the next second sentence. The information processing device inputs the sentence vector of the sentence of the article to be processed into the generated machine learning model, predicts the sentence vector of the next sentence, and based on the predicted sentence vector, detects the sentence with input errors from the article to be processed. That is, it is possible to detect a sentence including input errors or the like and having an inappropriate sentence vector from each sentence included in the article to be processed.

[0042] In addition, the information processing device may also be in Figure 2 In the process, when it is determined that the sentence "ペンギンは撮り" is a sentence with an inappropriate sentence vector, based on the sentence vector SV1-2 predicted by the machine learning model 50, a sentence with an appropriate sentence vector "ペンギンは鳥。" is retrieved from a DB (Data Base) or the like and output as a correct correction candidate to the display device (hereinafter referred to as appropriateness).

[0043] Moreover, the information processing device can also cause another machine learning model that has learned the order of the vectors of word units to calculate the word vectors of the multiple words "ペンギン", "は", and "撮り。" that make up the sentence "ペンギンは撮り" whose sentence vector is detected as inappropriate, and appropriate the input error of the deviated word "撮り。".

[0044] Next, a structural example of the information processing device that executes Figure 1 and Figure 2 the processing described in Figure 3 is a functional block diagram showing the structure of the information processing device according to the present embodiment. As Figure 3 shown, the information processing device 100 includes a communication unit 110, an input unit 120, a display unit 130, a storage unit 140, and a control unit 150.

[0045] The communication unit 110 is connected to an external device or the like by wire or wirelessly, and transmits and receives information to and from the external device or the like. For example, the communication unit 110 is implemented by a NIC (Network Interface Card) or the like. The communication unit 110 can also be connected to a network (not shown).

[0046] The input unit 120 is an input device that inputs various information into the information processing device 100. The input unit 120 corresponds to a keyboard, a mouse, a touch panel, etc. For example, a user can also operate the input unit 120 to input data such as an article.

[0047] The display unit 130 is a display device that displays the information output from the control unit 150. The display unit 130 corresponds to a liquid crystal display, an organic EL (Electro Luminescence) display, a touch panel, etc. For example, a sentence with an input error is displayed on the display unit 130.

[0048] The storage unit 140 includes a machine learning model 50, teacher data 141, and a word vector dictionary 142. The storage unit 140 is implemented by, for example, a semiconductor memory element such as a RAM (Random Access Memory), a flash memory, or a storage device such as a hard disk or an optical disc.

[0049] The machine learning model 50 is Figure 1 the NN such as BERT, Next Sentence Prediction, Transformers, etc. described in

[0050] The teacher data 141 is Figure 1The teacher data 141 described in . The article included in the teacher data 141 includes a plurality of sentences. The plurality of sentences have a predetermined relationship between the preceding and following sentences. Each sentence is a sentence set in advance based on the syllogism method of induction, deduction, etc.

[0051] The word vector dictionary 142 is a table that defines codes assigned to words and word vectors. Figure 4 is a diagram showing an example of the data structure of a word vector dictionary. Figure 4 As shown, the word vector dictionary 142 has codes, words, and word vectors (1) to (7). Codes are codes assigned to words. Words are words included in a character string. Word vectors (1) to (7) are vectors assigned to words. The nth component of a word vector is expressed as word vector (n) (n = 1 to 7).

[0052] The DB 143 has various articles. The articles include a plurality of sentences, and each sentence includes a plurality of words. The DB 143 may also include articles included in the teacher data 141 .

[0053] The sentence inversion index 144 associates a sentence vector with a position pointer. The position pointer indicates the position of the DB 143 where the sentence corresponding to the sentence vector exists.

[0054] return Figure 3 The control unit 150 includes a preprocessing unit 151, a learning unit 152, and an analyzing unit 153. The control unit 150 is implemented, for example, by a CPU (Central Processing Unit) or an MPU (Micro Processing Unit). In addition, the control unit 150 may also be implemented by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

[0055] The preprocessing unit 151 performs various preprocessing operations. For example, the preprocessing unit 151 obtains unprocessed sentences from the DB 143 and calculates sentence vectors of the sentences. The preprocessing unit 151 sets the relationship between the calculated sentence vector and the position pointer of the sentence corresponding to the sentence vector as the sentence inversion index 144.

[0056] An example of a process in which the preprocessing unit 151 calculates a sentence vector of a sentence will be described. Figure 5A as well as Figure 5BThis is a figure for explaining the processing of calculating sentence vectors. Here, the case of calculating the sentence vector of the sentence "A horse likes carrots." is explained. The preprocessing unit 151 decomposes the sentence "A horse likes carrots." into multiple words by performing morpheme analysis. "△ (space)" is assigned to each decomposed word. For example, sentence 1 "A horse likes carrots." is divided into "horse △", "is △", "carrots △", "likes △", "it △", "is △", ". △".

[0057] The preprocessing unit 151 determines the code corresponding to each word by comparing each of the segmented words with the word vector dictionary 45, and replaces it with the word. For example, each word "horse △", "is △", "carrots △", "likes △", "it △", "is △", ". △" is replaced with "C1", "C2", "C3", "C4", "C5", "C6", "C7" respectively.

[0058] Transfer to Figure 5B The explanation of. The preprocessing unit 151 determines the word vectors (1) to (7) assigned to the codes based on the word vector dictionary 45 and each code. For example, the word vectors (1) to (7) of the code "C1" are set as wv1-1 to 1-7. The word vectors (1) to (7) of the code "C2" are set as wv2-1 to 2-7. The word vectors (1) to (7) of the code "C3" are set as wv3-1 to 3-7.

[0059] The word vectors (1) to (7) of the code "C4" are set as wv4-1 to 4-7. The word vectors (1) to (7) of the code "C5" are set as wv5-1 to 5-7. The word vectors (1) to (7) of the code "C6" are set as wv6-1 to 6-7. The word vectors (1) to (7) of the code "C7" are set as wv7-1 to 7-7.

[0060] The preprocessing unit 151 calculates the sentence vector SV1 of the sentence by accumulating the word vectors for each element. For example, the preprocessing unit 151 calculates the first component "SV1-1" of the sentence vector SV1 by accumulating wv1-1 to 7-1 that become each word vector (1). The preprocessing unit 151 calculates the second component "SV1-2" of the sentence vector SV1 by accumulating wv1-2 to 7-2 that become each word vector (2). The third component "SV1-3" of the sentence vector SV1 is calculated by accumulating wv1-3 to 7-3 that become each word vector (3).

[0061] The preprocessing unit 151 calculates the fourth component "SV1-4" of the sentence vector SV1 by accumulating wv1-4 to 7-4, which become each word vector (4). The preprocessing unit 151 calculates the fifth component "SV1-5" of the sentence vector SV1 by accumulating wv1-5 to 7-5, which become each word vector (5). The preprocessing unit 151 calculates the sixth component "SV1-6" of the sentence vector SV1 by accumulating wv1-6 to 7-6, which become each word vector (6). The preprocessing unit 151 calculates the seventh component "SV1-7" of the sentence vector SV1 by accumulating wv1-7 to 7-7, which become each word vector (7).

[0062] The preprocessing unit 151 repeatedly executes the above-described processing on each sentence of other texts included in the DB 143 , thereby calculating a sentence vector for each sentence.

[0063] The preprocessing unit 151 generates a sentence inversion index 144 by establishing a correspondence between the calculated sentence vector of each sentence and the position pointer of the DB 143. Figure 6 The data structure shown has a sentence inversion index 144 . Figure 6 is a diagram for explaining the process of generating a sentence inverted index. Figure 6 As shown, the preprocessing unit 151 may establish correspondence between the sentence vector and a plurality of record pointers and a plurality of position pointers, and establish correspondence between each record pointer and position pointer and each sentence in the DB 143 .

[0064] return Figure 3 The learning unit 152 performs Figure 1 The learning phase described in the above description is used to generate a machine learning model 50, which, when a sentence vector of a first sentence is input, predicts a sentence vector of a second sentence following the first sentence.

[0065] For example, the learning unit 152 calculates the sentence vector of each sentence contained in the article of the teacher data 141 and sequentially inputs the calculated sentence vector into the machine learning model 50, thereby performing the learning of the machine learning model 50. Other processing of the learning unit 152 is similar to Figure 1 The process of calculating the sentence vector of a sentence by the learning unit 152 is the same as the process of calculating the sentence vector of a sentence by the preprocessing unit 151.

[0066] The analysis unit 153 executes Figure 2 The processing of the analysis stage described in is performed to detect sentences whose sentence vectors are inappropriate from the sentences contained in the article to be processed.

[0067] For example, when the analysis unit 153 receives the article 20 to be processed, it calculates the sentence vector of the sentence included in the article 20. The analysis unit 153 determines the sentence included in the article 20 based on the period "." included in the article 20. The process of calculating the sentence vector of the sentence by the analysis unit 153 is the same as the process of calculating the sentence vector of the sentence by the preprocessing unit 151. It is expressed as the sentence vector "SVn" (n = 0 to M) of the nth sentence from the beginning of the article 20.

[0068] The analysis unit 153 inputs the sentence vector SVn to the trained machine learning model 50, and predicts the sentence vector SVn+1' of the n+1th sentence from the beginning of the article 20. The analysis unit 153 calculates the cosine similarity between the sentence vector SVn+1' predicted by the machine learning model 50 and the sentence vector SVn+1.

[0069] When the cosine similarity between the sentence vector SVn+1' and the sentence vector SVn+1 is greater than the threshold, the analysis unit 153 determines that the n+1th sentence from the beginning is an appropriate sentence. On the other hand, when the cosine similarity between the sentence vector SVn+1' and the sentence vector SVn+1 is greater than the threshold, the analysis unit 153 determines that the n+1th sentence from the beginning is an inappropriate sentence vector.

[0070] When the analysis unit 153 determines that the sentence vector is inappropriate, it compares the sentence vector SVn+1' with the sentence inversion index 144 to determine the position pointer of the sentence corresponding to the sentence vector SVn+1'. Based on the position pointer, the analysis unit 153 retrieves the sentence corresponding to the sentence vector SVn+1' from the DB 143. The analysis unit 153 displays the sentence with inappropriate sentence vector in correspondence with the retrieved sentence on the display unit 130.

[0071] The analyzing unit 153 may compare the sentences with inappropriate sentence vectors and the retrieved sentences in units of words, detect incorrectly input words from the sentences with inappropriate sentence vectors, and display the detected words.

[0072] Next, an example of a processing procedure of the information processing device 100 according to the present embodiment will be described. Figure 7 1 is a flowchart showing the processing sequence of the learning phase of the information processing device involved in this embodiment. Figure 7 As shown, the learning unit 152 of the information processing device 100 selects unselected articles from the teacher data 141 (step S101).

[0073] The learning unit 152 calculates the sentence vector of each sentence included in the selected article, and generates a sentence inversion index that establishes a correspondence between the sentence vector, the DB record, and the position of the sentence (step S102). The learning unit 152 performs learning by sequentially inputting the sentence vector of the first sentence included in the selected article into the machine learning model 50 (step S103).

[0074] When the learning unit 152 continues the learning (step S104, Yes), the process proceeds to step S101. On the other hand, when the learning unit 152 does not continue the learning (step S104, No), the process of the learning phase ends.

[0075] Figure 8 FIG. 4 is a flowchart showing the processing sequence of the analysis phase of the information processing device involved in this embodiment. Figure 8 As shown, the analysis unit 153 of the information processing device 100 receives input of a text to be processed (step S201).

[0076] The analysis unit 153 calculates the sentence vector of each sentence included in the input text (step S202 ). The analysis unit 153 sets n to an initial value (step S203 ).

[0077] The analysis unit 153 inputs the sentence vector SVn of the nth sentence among the plurality of sentences included in the article to the machine learning model 50, and predicts the sentence vector SVn+1′ of the n+1th sentence (step S204).

[0078] The analysis unit 153 calculates the cosine similarity between the sentence vector SVn+1 of the n+1th sentence among the plurality of sentences included in the text and the sentence vector SVn+1′ of the predicted sentence (step S205 ).

[0079] When the cosine similarity is equal to or greater than the threshold value (step S206 , Yes), the analysis unit 153 proceeds to step S210 .

[0080] On the other hand, when the cosine similarity is not greater than the threshold value (step S206, No), the analysis unit 153 detects the n+1th sentence as a sentence with an inappropriate sentence vector (step S207). The analysis unit 153 detects the sentence corresponding to the sentence vector SVn+1' from the DB 143 based on the predicted sentence vector SVn+1' and the sentence inversion index 144 (step S208).

[0081] The analyzing unit 153 displays the sentences with inappropriate sentence vectors and the sentences detected from the DB 143 on the display unit 130 (step S209 ).

[0082] The processing after step S210 is described. When n is greater than L (step S210, yes), the analysis unit 153 ends the processing. L is the number of sentences contained in the article to be processed. When n is not greater than L (step S210, no), the analysis unit 153 updates n by adding 1 to n (step S211), and transfers to step S204.

[0083] Next, the effect of the information processing device 100 involved in this embodiment is described. The information processing device 100 generates a machine learning model 50 by sequentially inputting the vectors of each sentence of the article contained in the teacher data 141 into the machine learning model 50. When the sentence vector of a first sentence is input, the machine learning model predicts the sentence vector of the second sentence next to the first sentence. The information processing device 100 inputs the sentence vector of the sentence of the article to be processed into the generated machine learning model 50, predicts the sentence vector of the next sentence, and detects sentences with inappropriate sentence vectors from the article to be processed based on the predicted sentence vector. In addition, words such as input errors can be appropriated from the inappropriate sentence.

[0084] The information processing device 100 detects sentences with inappropriate sentence vectors and corrects input errors, etc. based on the cosine similarity between the sentence vector of the next sentence predicted by the machine learning model 50 and the sentence vector of the next sentence of the sentence included in the article to be processed. In this way, it is possible to suppress the computational cost, detect sentences with inappropriate sentence vectors, and correct input errors, etc.

[0085] The information processing device 100 sequentially inputs the vectors of a plurality of sentences whose arrangement order is determined based on induction or deduction into a machine learning model for training, thereby being able to predict the next sentence of the target sentence based on induction or deduction.

[0086] When the information processing device 100 determines that a sentence is a correction target, it searches for the sentence to be corrected based on the vector predicted by the machine learning model 50. In this way, the corrected sentence can be notified.

[0087] The above-described processing contents of the information processing device 100 are merely examples, and the information processing device 100 may also execute other processing. Hereinafter, other processing of the information processing device 100 will be described.

[0088] Fig. 9 as well as Fig.10This is a diagram for explaining other processing of the information processing device. The above-mentioned information processing device 100 causes the machine learning model 50 to learn the order of the vectors of the sentence based on the syllogism method, but it is also possible to replace the sentence vectors with the protein sequence and learn the order of the vectors of the primary structure of the protein composed of multiple amino acid sequences equivalent to words. In the following description, the continuous amino acid sequence of the protein is expressed as the "basic structure" and the primary structure of the protein is expressed as the "primary structure".

[0089] right Fig. 9 In the learning phase, the information processing device 100 uses the sequences 20 a and 20 b of a plurality of proteins included in the teacher data 241 to perform learning of the machine learning model 50 .

[0090] For example, the sequence 20a includes the primary structures "α primary structure", "β primary structure", ..., and "γ primary structure", and the sequence 20b includes the primary structures "Δ primary structure", "ε primary structure", ..., and "ζ primary structure".

[0091] The information processing device 100 uses a vector dictionary of the basic structure of the protein that establishes a correspondence between the basic structure and the vector to determine the vector of each primary structure. For example, the vector of the primary structure "α primary structure" composed of multiple basic structures is set to "V20-1", the vector of the primary structure "β primary structure" is set to "V20-2", and the vector of the primary structure "γ primary structure" is set to "V20-3". The vector of the primary structure "Δ primary structure" is set to "V21-1", the vector of the primary structure "ε primary structure" is set to "V21-2", and the vector of the primary structure "ζ primary structure" is set to "V21-3". The vector of each primary structure is calculated based on the vectors of each basic structure of the multiple basic structures that constitute the primary structure.

[0092] The information processing device 100 repeatedly performs the process of sequentially inputting the vector of the primary structure contained in the protein sequence to the machine learning model 50. For example, the information processing device inputs vectors in the order of vectors "V20-1", "V20-2", ..., "V20-3" to the machine learning model 50. The information processing device inputs vectors in the order of vectors "V21-1", "V21-2", ..., "V21-3" to the machine learning model 50.

[0093] The information processing device 100 executes the above-mentioned processing in the learning phase to generate a machine learning model 50 that predicts the vector of the next primary structure of a certain primary structure when a vector of a certain primary structure is input.

[0094] right Fig.10In the analysis stage, the sequence of the protein to be processed is set to sequence 25. Sequence 25 includes primary structures "α primary structure", "η primary structure", ..., "γ primary structure" in order from the beginning.

[0095] The information processing device 100 calculates the vector "V20-1" of the primary structure "α primary structure" and inputs the calculated vector "V20-1" into the machine learning model 50, thereby predicting the vector of the next primary structure of the primary structure "α primary structure". Fig.10 In the example shown, the machine learning model 50 predicts "V20-2" as the vector of the next primary structure of the primary structure "α primary structure".

[0096] The information processing apparatus 100 calculates a vector “V22” which is a primary structure included in the sequence 25 and is the next “η primary structure” of the primary structure “α primary structure”.

[0097] The information processing device 100 calculates the cosine similarity between the vector "V20-2" of the next primary structure predicted by the machine learning model 50 and the vector "V22" of the next "η primary structure" which is the primary structure contained in the sequence 25 and is the basic structure "α primary structure".

[0098] When the cosine similarity is greater than the threshold, the information processing device determines that the primary structure included in the sequence 25 and the next "n primary structure" of the primary structure "α primary structure" is the correct primary structure. On the other hand, when the cosine similarity is less than the threshold, the information processing device determines that the primary structure included in the sequence 25 and the next "n primary structure" of the primary structure "α primary structure" is an inappropriate primary structure, and makes the mutation of the basic structure included in the primary structure "n primary structure" appropriate.

[0099] By executing the information processing device 100 Fig. 9 as well as Fig.10 The processing shown can detect a primary structure having an inappropriate primary structure vector from multiple primary structures contained in the sequence of a protein, and can make the basic structure with mutations etc. appropriate. Thus, it is possible to detect a protein primary structure having mutations etc. (SNPs are a representative example) generated in a receptor composed of multiple protein primary structures. Furthermore, by performing machine learning on multiple protein primary structures constituting the receptor and a single or multiple protein primary structures binding to the receptor according to the binding order, it is possible to predict the vector of the protein primary structure of the ligand binding to the receptor. Thus, it is possible to support the improvement of ligands having novel protein primary structure vectors that have excellent efficacy and suppressed side effects, similar to ligands that have been commercialized as biopharmaceuticals.

[0100] Next, an example of a hardware configuration of a computer that realizes the same functions as the information processing device 100 shown in the above embodiment will be described. Fig.11 This is a diagram showing an example of the hardware configuration of a computer that realizes the same function as the information processing device of the embodiment.

[0101] like Fig.11 As shown, the computer 300 has a CPU 301 that performs various calculations, an input device 302 that receives data input from a user, and a display 303. In addition, the computer 300 has a communication device 304 and an interface device 305 that perform data transmission and reception with external devices via a wired or wireless network. In addition, the computer 300 has a RAM 306 and a hard disk device 307 that temporarily store various information. In addition, each device 301 to 307 is connected to a bus 308.

[0102] The hard disk device 307 has a preprocessing program 307 a , a learning program 307 b , and an analysis program 307 c . The CPU 301 reads out each of the programs 307 a to 307 c and expands them in the RAM 306 .

[0103] The preprocessing program 307a functions as the preprocessing process 306a. The learning program 307b functions as the learning process 306b. The analysis program 307c functions as the analysis process 306c.

[0104] The processing of the preprocessing process 306a corresponds to the processing of the preprocessing unit 151. The processing of the learning process 306b corresponds to the processing of the learning unit 152. The processing of the analysis process 306c corresponds to the processing of the analysis unit 153.

[0105] In addition, each program 307a to 307c may not necessarily be initially stored in the hard disk device 307. For example, each program may be stored in a "removable physical medium" such as a floppy disk (FD), CD-ROM, DVD, optical magnetic disk, IC card, etc. inserted into the computer 300. Furthermore, the computer 300 may read out each program 307a to 307c and execute it.

[0106] Description of Reference Numerals

[0107] 50…machine learning model; 100…information processing device; 110…communication unit; 120…input unit; 130…display unit; 140…storage unit; 141…teacher data; 142…word vector dictionary; 143…DB; 144…sentence inversion index; 150…control unit; 151…preprocessing unit; 152…learning unit; 153…analysis unit.

Claims

1. An information processing program, characterized in that: Make the computer execute the following processing: Calculate vectors of a plurality of consecutive sentences respectively, which are the plurality of sentences having a relationship between the preceding and following sentences; Generating the machine learning model by sequentially inputting the vectors of the plurality of sentences into the machine learning model and training the model, wherein when a vector of a certain sentence is input into the machine learning model, the machine learning model predicts a sentence vector of a sentence that is input next to the certain sentence; Calculating a vector of a first sentence and a vector of a second sentence following the first sentence; and By inputting the vector of the first sentence into the machine learning model, a vector of a sentence predicted to follow the first sentence is calculated, and it is determined whether the vector of the second sentence is appropriate.

2. The information processing program according to claim 1, characterized in that In the determination process, whether the vector of the second sentence is appropriate is determined based on the cosine similarity between the vector predicted by inputting the vector of the first sentence into the machine learning model and the vector of the second sentence.

3. The information processing program according to claim 2, characterized in that The multiple consecutive sentences are multiple sentences whose arrangement order is determined based on induction or deduction. In the process of generating the machine learning model, the vectors of the multiple sentences whose arrangement order is determined based on induction or deduction are sequentially input into the machine learning model and trained.

4. The information processing program according to claim 1, characterized in that The computer is caused to further perform the following processing: when it is determined that the vector of the second sentence is inappropriate, a vector of a sentence predicted to follow the first sentence is calculated by inputting the vector of the first sentence into the machine learning model, and an appropriate sentence is recommended based on the vector of the sentence predicted to follow the calculated first sentence in order to retrieve a sentence similar to the calculated vector and suggest it as a candidate for an appropriate sentence.

5. An information processing method, characterized in that: The following processing is performed by the computer: Calculate vectors of a plurality of consecutive sentences respectively, which are the plurality of sentences having a relationship between the preceding and following sentences; Generating the machine learning model by sequentially inputting the vectors of the plurality of sentences into the machine learning model and training the model, wherein when a vector of a certain sentence is input into the machine learning model, the machine learning model predicts a sentence vector of a sentence that is input next to the certain sentence; Calculating a vector of a first sentence and a vector of a second sentence following the first sentence; and By inputting the vector of the first sentence into the machine learning model, a vector of a sentence predicted to follow the first sentence is calculated, and it is determined whether the vector of the second sentence is appropriate.

6. The information processing method according to claim 5, characterized in that: In the determination process, whether the vector of the second sentence is appropriate is determined based on the cosine similarity between the vector predicted by inputting the vector of the first sentence into the machine learning model and the vector of the second sentence.

7. The information processing method according to claim 6, characterized in that: The multiple consecutive sentences are multiple sentences whose arrangement order is determined based on induction or deduction. In the process of generating the machine learning model, the vectors of the multiple sentences whose arrangement order is determined based on induction or deduction are sequentially input into the machine learning model and trained.

8. The information processing method according to claim 6, characterized in that: The computer is caused to further perform the following processing: when it is determined that the vector of the second sentence is inappropriate, a vector of a sentence predicted to follow the first sentence is calculated by inputting the vector of the first sentence into the machine learning model, and an appropriate sentence is recommended based on the vector of the sentence predicted to follow the calculated first sentence in order to retrieve a sentence similar to the calculated vector and suggest it as a candidate for an appropriate sentence.

9. An information processing device, characterized in that: A control unit is provided, which performs the following processing: Calculate vectors of a plurality of consecutive sentences respectively, which are the plurality of sentences having a relationship between the preceding and following sentences; Generating the machine learning model by sequentially inputting the vectors of the plurality of sentences into the machine learning model and training the model, wherein when a vector of a certain sentence is input into the machine learning model, the machine learning model predicts a sentence vector of a sentence that is input next to the certain sentence; Calculating a vector of a first sentence and a vector of a second sentence following the first sentence; and By inputting the vector of the first sentence into the machine learning model, a vector of a sentence predicted to follow the first sentence is calculated, and it is determined whether the vector of the second sentence is appropriate.

10. The information processing device according to claim 9, characterized in that: In the determination process performed by the control unit, whether the vector of the second sentence is appropriate is determined based on the cosine similarity between the vector predicted by inputting the vector of the first sentence into the machine learning model and the vector of the second sentence.

11. The information processing device according to claim 10, characterized in that: The multiple consecutive sentences are multiple sentences whose arrangement order is determined based on induction or deduction. In the process of generating the machine learning model performed by the control unit, the vectors of the multiple sentences whose arrangement order is determined based on induction or deduction are sequentially input into the machine learning model and trained.

12. The information processing device according to claim 9, characterized in that: The control unit causes the computer to also perform the following processing: when it is determined that the vector of the second sentence is inappropriate, a vector of a sentence predicted to follow the first sentence is calculated by inputting the vector of the first sentence into the machine learning model, and an appropriate sentence is recommended based on the vector of the sentence predicted to follow the calculated first sentence in order to retrieve a sentence similar to the calculated vector and suggest it as a candidate for an appropriate sentence.

Images