JP2020144844A - System and method for processing information - Google Patents

Abstract

To provide a system and method for processing information, which improve efficiency in creating and/or checking contract documents.SOLUTION: An information processing system provided herein consists of terminals and a server connected via a communication network, the server comprising: a storage unit storing analyzed relationships between vectors obtained by converting contract terms and term type identification information indicative of types of the terms, or term metadata for the terms; a conversion unit for converting terms of interest of a contract of interest into vectors; and a prediction unit configured to use the converted vectors as an input to predict and output term type identification information for identifying types of the terms of interest, or term metadata for the terms of interest, based on the analyzed relationships.SELECTED DRAWING: Figure 4

Description

The present invention relates to an information processing system and an information processing method.

In business involving transactions, contracts that describe mutual transaction conditions are important documents that function as evidence at the time of contract conclusion, and the creation of contracts is a very important business. As a technique related to contract creation, for example, in Patent Document 1, all contract patterns assumed in advance for each contract type in the computer database, and cases for each contract pattern are modified for each individual case. By accumulating the cases where the parts to be to be displayed, the assumed risks, etc. are accumulated, the contract document creator selects a contract type similar to the case from the contract type on the terminal device of the contract document creation system, and sets the selected contract type. The corresponding contract patterns are displayed in sequence, the required case is selected for each displayed contract pattern, the selected pattern and the risk along the case are displayed as necessary, and the entire selection is completed. A contract document creation system that creates a contract document by modifying the revised part to a contract document and words and phrases suitable for individual cases is disclosed.

Japanese Unexamined Patent Publication No. 2006-65421

Matteo Pagliardini, Prakhar Gupta, Martin Jaggi, "Unsupervised Learning of Sentence Embeddings using Compositional n-Gram Features", NAACL 2018 --Conference of the North American Chapter of the Association for Computational Linguistics, pages 528-540

In the technique described in Patent Document 1, it is assumed that all contract patterns assumed in advance are prepared and a model contract selected from the contracts is used. However, in the actual contract, not only the case where the contract of the template prepared by the company in advance is used, but also the case where the contract is prepared by the company for the first time or the contract presented by the other party is used. In these cases, if the person in charge of the contract is unfamiliar with the type of contract, there is a problem that it is difficult to grasp what kind of clause the clause is. Another problem is that if the person in charge of the contract is unfamiliar with that type of contract, it is difficult to create and / or check the contract without knowing whether each clause is advantageous or disadvantageous to him / her. there were. Due to such a problem, there is a problem that the efficiency of contract creation and / or check is inefficient.

The present invention has been made in view of the above problems, and an object of the present invention is to provide an information processing system and an information processing method capable of improving the efficiency of contract creation and / or checking.

The information processing system according to the first aspect of the present invention stores the analyzed relationship between the vector in which the clause of the contract is converted and the clause type identification information indicating the type of the clause or the clause metadata for the clause. Based on the stored storage, the conversion unit that converts the target clause of the target contract into a vector, and the analyzed relationship, the converted vector is used as input.
It is provided with a clause type identification information for identifying the type of the target clause or a prediction unit for predicting and outputting clause metadata for the target clause.

According to this configuration, by assigning the clause type identification information to the target clause, the user can grasp the type of the target clause, so that the user can grasp what kind of clause the contract clause is. Can be facilitated. This can improve the efficiency of contract creation and / or checking.
Alternatively, according to this configuration, if the clause metadata suggests which side of the contract is the disadvantageous clause, the target user may change the clause that is disadvantageous to him or herself so that it is not disadvantageous. New clauses can be added so as not to be disadvantageous. In addition, if the clause metadata suggests which side of the contract is the favorable clause, the target user can create and / or check the contract while grasping the clause that is advantageous to him / her. This can improve the efficiency of contract creation and / or checking.

The information processing system according to the second aspect of the present invention is the information processing system according to the first aspect, and the analyzed relationship stored in the storage is a vector of any provision included in the standard contract. It is a classification model machine-learned as teacher data that inputs a vector of clauses that are not similar to the setting criteria or more and identifies the clause type assigned by the user to the clause and outputs clause type identification information. The unit inputs the converted vector into the classification model and outputs the clause type identification information for identifying the type of the target clause.

According to this configuration, the accuracy of discriminating the type of clause can be improved.

The information processing system according to the third aspect of the present invention is the information processing system according to the first aspect, and the analyzed relationships stored in the storage correspond to each of the provisions included in the standard contract. It is a set of a vector and clause type identification information indicating the type of the clause, and the prediction unit uses the converted vector as a vector of each clause included in the standard contract stored in the storage. As a result of comparison, the clause type identification information associated in the storage is output to a vector whose similarity satisfies the standard.

This structure allows the subject clause to be compared to the clause contained in the standard contract.
Can determine the type of clause in question.

The information processing system according to the fourth aspect of the present invention is the information processing system according to the third aspect, and the storage is attached to the vector corresponding to each clause included in the standard contract and the clause. The type of contract is further associated and stored with respect to the set with the clause type identification information, and further includes an acquisition unit for acquiring the type of the target contract, and the prediction unit is after the conversion. Is compared with the vector of each clause contained in a standard contract of the same type as the subject contract type acquired by the acquisition unit.

According to this configuration, the target clause can be compared with the clause contained in the standard contract of the same type as the target contract type, and the accuracy of determining the type of the target clause can be improved.

The information processing system according to the fifth aspect of the present invention is the information processing system according to the third or fourth aspect, and the conversion unit converts each of the provisions included in the target contract into a vector, and the above-mentioned The prediction unit compares each of the converted vectors with the vector of each clause included in the standard contract of the same type as the target contract, and as a result of the comparison by the prediction unit, among the converted vectors, among the converted vectors. It is further provided with an output unit that outputs a clause corresponding to a vector having no similarity above the setting standard with the vector of any clause included in the standard contract as a dissimilar clause.

With this configuration, the user can grasp the provisions that are not in the standard contract, so
It is possible to streamline the creation and / or check of contracts.

The information processing system according to the sixth aspect of the present invention is the information processing system according to the fifth aspect, and the user inputs the vector of the dissimilar clause output by the output unit to the clause. An update unit that updates the classification model by machine learning using the teacher data that outputs the clause type identification information that identifies the type of the assigned clause is further provided, and the prediction unit is added to the updated classification model. The converted vector is input, and the clause type identification information that identifies the type of the target clause is output.

According to this configuration, the accuracy of assigning the correct clause type identification information to the clause can be improved by using the updated classification model.

The information processing system according to the seventh aspect of the present invention is the information processing system according to the third or fourth aspect, and the conversion unit converts each of the provisions included in the target contract into a vector, and the above-mentioned The prediction unit compares each of the converted vectors with each vector of the provisions included in the standard contract of the same type as the target contract, and as a result of the comparison by the prediction unit, the provisions included in the standard contract. Of the vectors of, the information corresponding to the clause type identification information indicating the type of clause corresponding to the vector having no similarity equal to or higher than the setting standard with the converted vector is insufficient in the target contract. It also has an output unit that outputs as a clause.

With this configuration, the user can grasp the provisions that are not in the standard contract, so
It is possible to streamline the creation and / or check of contracts.

The information processing system according to the eighth aspect of the present invention is the information processing system according to any one of the first to seventh aspects, and the storage has one or more provisions with respect to the clause type identification information. Clause type that identifies the type of the particular clause if the data and / or clause metadata is associated and stored and accepts the user's particular operation on the particular clause contained in the subject agreement. The same clause type as the identification information The clause data output unit for outputting the clause data and / or the clause metadata associated with the identification information in the storage is further provided.

According to this configuration, it is possible to refer to other clauses of the same type as the clause selected by the user, so that the creation of the contract can be made more efficient.

The information processing system according to the ninth aspect of the present invention is the information processing system according to any one of the first to eighth aspects, and the conversion unit changes the provisions of the target contract from Japanese to a foreign language. Translate and convert the translated foreign language clause into a vector.

According to this configuration, it is converted to a vector after being converted to a foreign language, so after absorbing the notational fluctuations in Japanese and converting to a vector, those with similar meanings are converted into vectors with higher similarity, and the meaning is It is possible to convert distant objects into vectors with lower similarity.

The information processing system according to the tenth aspect of the present invention is the information processing system according to the ninth aspect, and the foreign language is English.

According to this configuration, by translating into English and vectorizing from English, English is less ambiguous and less syntactically fluctuating than Japanese, so when vectorized, it is converted from Japanese to vector. It is possible to convert those having similar meanings into a vector having a higher degree of similarity, and those having a distant meaning into a vector having a lower degree of similarity.

The information processing system according to the eleventh aspect of the present invention is an information processing system according to any one of the first to tenth aspects, further comprising a dividing portion for dividing the target contract into clauses. The conversion unit converts the clause after the division by the division unit into a vector.

According to this configuration, the target contract is divided into clauses and the divided clauses are converted into vectors, so that the clause type identification information can be assigned to each clause of the target contract.

The information processing system according to the twelfth aspect of the present invention is the information processing system according to the first aspect, and the storage is a set of a user classification and a vector of standard clauses advantageous or disadvantageous for the user classification. And the analyzed relationship with the clause metadata are stored, and the prediction unit inputs the clause metadata by inputting the classification of the target user and the vector of the target clause based on the analyzed relationship. Predict and output.

According to this configuration, if the clause is disadvantageous in the target user's classification, the clause metadata suggesting to change or add the clause is obtained, so that the target user does not disadvantage the clause that is disadvantageous to himself / herself. You can change it to, or add a new clause so that it does not disadvantage you. In addition, in the case of a clause that is advantageous in the classification of the target user, clause metadata suggesting that the clause is advantageous can be obtained, so that the target user can create a contract while grasping the clause that is advantageous to him / herself. / Or can be checked. This can improve the efficiency of contract creation and / or checking.

The information processing method according to the thirteenth aspect of the present invention has a storage that stores the vector in which the clause is converted and the clause type identification information representing the type of the clause or the analyzed relationship between the clause metadata for the clause. An information processing method executed by a stored information processing system, in which the converted vector is used as an input based on the step of converting the target clause of the target contract into a vector and the analyzed relationship. , The step of predicting and outputting the clause type identification information for identifying the type of the subject clause or the clause metadata for the subject clause.

According to this configuration, by assigning the clause type identification information to the target clause, the user can grasp the type of the target clause, so that the user can grasp what kind of clause the contract clause is. Can be facilitated. This can improve the efficiency of contract creation and / or checking. Alternatively, according to this configuration, clause metadata can be obtained, so that the target user can change the clause that is disadvantageous to himself / herself so as not to be disadvantageous, or add a new clause so as not to be disadvantageous to himself / herself. can do. In addition, since clause metadata suggesting that the clause is advantageous can be obtained, the target user can create and / or check the contract while grasping the clause that is advantageous to him / her. This can improve the efficiency of contract creation and / or checking.

According to one aspect of the present invention, by assigning the clause type identification information to the target clause, the user can grasp the type of the target clause, so that what kind of clause the contract clause is. It can be facilitated for the user to grasp. This will create a contract and /
Alternatively, the efficiency of the check can be improved.
Alternatively, according to one aspect of the invention, if the clause metadata suggests which side of the contract is the disadvantageous clause, the subject user may modify the clause that is disadvantageous to him or her so that it is not disadvantageous. , You can add new clauses so as not to be disadvantageous to you. Also, if the clause metadata suggests which side of the contract is in favor of the clause,
The target user can create and / or check the contract while grasping the clauses that are advantageous to him / her. This can improve the efficiency of contract creation and / or checking.

It is a schematic diagram explaining the learning method of the classification model of the clause which concerns on this embodiment. It is a schematic block diagram of the information processing system which concerns on this embodiment. It is a schematic block diagram of the terminal which concerns on this embodiment. It is a schematic block diagram of the server which concerns on this embodiment. This is an example of a master table stored in the storage of the server according to the present embodiment. This is an example of a table stored in the storage of the server according to the present embodiment. It is a schematic diagram explaining one of the processing which concerns on this embodiment. It is a schematic diagram which shows an example of vectorization of a clause in the case where the number of sentences contained in a clause is three. This is the first example of the browser screen displayed on the terminal 1. FIG. 9 is an example of a browser screen when the “Search by this tag” button B1 is pressed. It is a flowchart which shows an example of the process when the prediction part assigns a clause type to a clause using the similarity in this embodiment. It is a flowchart which shows an example of the process which presents a clause type and a similar clause in this embodiment. It is a flowchart which shows an example of the processing when the prediction part assigns a clause type to a clause using a classification model in this embodiment.

Hereinafter, each embodiment will be described with reference to the drawings. However, more detailed explanation than necessary may be omitted. For example, detailed explanations of already well-known matters and duplicate explanations for substantially the same configuration may be omitted. This is to avoid unnecessary redundancy of the following description and to facilitate the understanding of those skilled in the art.

In this embodiment, in addition to or in place of the above problems, the following problems are solved. For example, when preparing a contract in-house for the first time or when the other party presents the contract,
There is a challenge to improve the efficiency of contract creation and / or checking. In order to improve the efficiency of contract creation and / or checking, among the target contracts, provisions that are not common in contracts of that type (for example, in the case of joint development contracts, in joint development contracts) If you can identify an uncommon clause, etc.), you can focus on that clause and check it. However, there is a problem that it is difficult to distinguish clauses that are not common in contracts of that type from the target contracts.

In addition, in order to improve the efficiency of contract creation and / or checking, the target contract has provisions that are insufficient for that type of contract (for example, if it is a joint development contract, joint development is used. If you can understand what kind of clause (such as the clause that stipulates the treatment of the intellectual property right that was born), you can add that clause. However, there is a problem that it is difficult to grasp what kind of clause is lacking in the target contract as a contract of that type.

FIG. 1 is a schematic diagram illustrating a learning method of a classification model of the clauses according to the present embodiment. Figure 1
As shown in, in the database of standard contracts, which are standard contracts, standard contracts are divided and stored for each clause.
(Step S1) First, for example, the server is the contract data DT, which is the data of the target contract.
Is used to divide the target contract into clause units. Then, for example, the server combines and clusters (groups) the data of each clause of the contract data DT and the data of each clause of the standard contract, and acquires the vector of each clause.

(Step S2) For example, the server is a vector V11 to V of each clause of the target contract.
For 16, the vectors V21 to V2 of each clause of the standard contract of the same type as the target contract
The similarity with 6 (eg, cosine similarity) is calculated. In the present embodiment, the similarity will be described as an example of the cosine degree.

Among the provisions of the target contract, the provisions whose similarity is equal to or higher than the setting standard (here, V12 to V15)
Exclude. For example, the vector V11 is left because the similarity with each of the vectors V21 to V26 does not exceed the setting standard. Similarly, for example, the vector V16 is left because the similarity with each of the vectors V21 to V26 does not exceed the setting standard. As a result, for example, the vectors V11 and V16 do not have a clause in the standard contract whose similarity is equal to or higher than the setting standard, so that the server suggests to the user as an uncommon clause.

On the other hand, for example, the server has a vector V21 to V26 of the clauses of the standard contract similar to the target contract for each of the clauses of the target contract whose similarity is equal to or higher than the setting standard (here, V12 to V15). The clause type corresponding to the clause of the vector having the highest similarity is assigned to the clause type of the vector. Specifically, for example, if the server is a vector V23 having a similarity equal to or higher than the setting standard and most similar to the clause V12, the server assigns the clause type corresponding to the vector V23 to the clause 12.

For example, the vector V21 is left because the similarity with each of the vectors V11 to V16 does not exceed the setting reference. Similarly, for example, the vector V21 is a vector V11 to 1.
Since none of the similarities with each of V16 exceeded the setting standard, it is left. As a result, for example, the vectors V21 and V26 did not have a clause in the target contract whose similarity was equal to or higher than the setting standard, and therefore, for example, the server suggested to the user that the clause was insufficient for that type of contract. To do.

(Step S3) The user assigns a clause type to each of the dissimilar clauses whose similarity is less than the setting standard (here, clauses corresponding to vectors V11 and V16, for example). For example, if the user reads the clause of vector V11 and determines that it is a clause of "confidentiality", the user operates a terminal to assign "confidentiality" to the clause of vector V11.

(Step S4) Subsequently, for example, the server uses the teacher data that inputs the vector of the dissimilar clause and outputs the clause type ID that identifies the clause type assigned by the user to the dissimilar clause, and performs machine learning. Update the classification model by doing.

(Step S5) When certain performance requirements are met, each type of clause is predicted using this classification model instead of similarity.

It is a schematic block diagram of the information processing system which concerns on this embodiment. As shown in FIG. 2, the information processing system S includes, as an example, a server 2 connected to terminals 1-1 to 1-N via a communication network CN. The information processing system S may or may not include terminals 1-1, ..., 1-N (N is a natural number).

The terminals 1-1 to 1-N are terminal devices used by different users, and are, for example, mobile phones such as multifunctional mobile phones (so-called smartphones), tablets, notebook computers, desktop personal computers, and the like. Terminals 1-1 to 1-N use, for example, a WEB browser to display information provided by the server 2.

The server 2 is an example of an information processing device, and provides information to terminals 1-1 to 1-N.
Hereinafter, terminals 1-1 to 1-N are collectively referred to as terminal 1.

FIG. 3 is a schematic configuration diagram of a terminal according to the present embodiment. As shown in FIG. 3, the terminal 1 includes, for example, an input interface 11, a communication circuit 12, a storage 13, a memory 14, an output interface 15, and a processor 16.
The input interface 11 receives an input from the user and outputs an input signal corresponding to the received input to the processor 16.
The communication circuit 12 is connected to the communication network CN and communicates with the server 2 connected to the communication network CN. This communication may be wired or wireless.

The storage 13 stores a program and various data for the processor 16 to read and execute.
The memory 14 temporarily holds data and programs. The memory 14 is a volatile memory, for example, a RAM (Random Access Memory).

The output interface 15 is connected to the display 17, and outputs information to the display 17 according to a command from the processor 16. As a result, the information is displayed on the display 17. The display 17 may be built in the terminal 1 instead of being externally attached to the terminal 1.
The processor 16 loads the program according to the present embodiment from the storage 13 into the memory 14, and executes a series of instructions included in the program to execute the read reception unit 16
1. Editing reception unit 162, clause data request unit 163, clause type request unit 164. Communication control unit 1
65, functions as an output control unit 166. Each process will be described later.

FIG. 4 is a schematic configuration diagram of a server according to the present embodiment. As shown in FIG. 4, the server 2 includes an input interface 21, a communication circuit 22, a storage 23, a memory 24, an output interface 25, and a processor 26.
The input interface 21 receives an input from the administrator of the server 2 and outputs an input signal corresponding to the received input to the processor 26.
The communication circuit 22 is connected to the communication network CN and communicates with terminals 1-1 to 1-N connected to the communication network CN. This communication may be wired or wireless.

The storage 23 stores a program and various data for the processor 26 to read and execute.
The memory 24 temporarily holds data and programs. The memory 24 is a volatile memory, for example, a RAM (Random Access Memory).
The output interface 25 is connected to an external device (for example, a display, and outputs a signal (for example, a video signal) to the external device according to a command from the processor 26. As a result, a video signal is input to the display, for example. Information is displayed.

The processor 26 loads the program from the storage 23 into the memory 24 and executes a series of instructions included in the program to execute the acquisition unit 261 and the division unit 262, the conversion unit 263, the prediction unit 264, and the clause data reading unit ( It functions as a clause data output unit) 265, a clause type reading unit 266, an output unit 267, an update unit 268, and a communication control unit 269. Each process will be described later.

FIG. 5 is an example of a master table stored in the storage of the server according to the present embodiment. As shown in FIG. 5, a set of a contract type ID and a clause name is stored in the contract type master table M1 stored in the storage 23 of the server 2 according to the present embodiment. Here, the contract type ID is an example of contract type identification information for identifying the type of contract.
As shown in FIG. 5, a set of a clause type ID and a clause name is stored in the clause type master table M2 stored in the storage 23 of the server 2 according to the present embodiment. Here, the clause type ID is an example of clause type identification information for identifying the clause type.

FIG. 6 is an example of a table stored in the storage of the server according to the present embodiment. The standard clause table T1 stores information about the clauses included in the standard contract prepared in advance. As shown in FIG. 6, in the standard clause table T1 stored in the storage 23 of the server 2 according to the present embodiment, a record of a set of a contract type ID, a clause type ID, a vector, and a clause is stored. It has been accumulated. A vector is a conversion of a clause, that is, a sentence group of clause units. The vector is, for example, a high-dimensional real vector. At the time of conversion, clauses with similar meanings are converted into close vectors.
In this way, in the storage 23, for the set of the vector corresponding to each clause included in the standard contract and the clause type identification information (here, the contract type ID) attached to the clause, the contract Types are further associated and stored.

In the clause table T2, the clauses included in the contract entered by the user are accumulated. As shown in FIG. 6, the clause table T2 stored in the storage 23 of the server 2 according to the present embodiment stores a record of a set of a contract type ID, a clause type ID, and a clause. ..

The storage 23 of the server 2 stores the analyzed relationship between the vector in which the terms of the contract are converted and the clause type identification information (here, the clause type ID) indicating the type of the clause.
In this embodiment, there are two types of analyzed relationships stored in the storage 23. One of the analyzed relationships stored in the storage 23 is the vector corresponding to each clause included in the standard contract and the clause type identification information indicating the type of the clause (here, the clause type ID).
). The other analyzed relationship stored in the storage 23 is to input the vector of the clause that is not similar to the vector of any clause contained in the standard contract more than the setting standard and assign it to the clause by the user. It is a classification model machine-learned as teacher data that outputs the clause type identification information that identifies the clause type.

FIG. 7 is a schematic diagram illustrating one of the processes according to the present embodiment. The reading reception unit 161
The contract data DT provided according to the operation by the user is read, and the communication control unit 165 (
(See FIG. 3) controls the communication circuit 12 to transmit the contract data DT to the server 2.
The editing reception unit 162 accepts editing for the contract data DT, and the output control unit 166
(See FIG. 3) controls the edited contract data DT to be displayed on the display 17.

The acquisition unit 261 acquires the type of the target contract. At that time, for example, the acquisition unit 261
The title included in the target contract may be acquired as the type of contract, or the type of contract may be determined from the title included in the target contract. Further, the user may specify the type of the target contract, and in that case, the acquisition unit 261 may acquire the type of the target contract received by the terminal 1 from the user.

The division unit 262 divides the target contract into clauses. In the process of this division, for example, the division unit 262 may extract the word "article", regard the "article" as a boundary, and divide it into each clause.

The conversion unit 263 converts the target clause of the target contract into a vector. Specifically, for example, the conversion unit 263 converts the clause after the division by the division unit 262 into a vector. This conversion will be described later.

Based on the above-analyzed relationship, the prediction unit 264 uses the converted vector as an input to predict and output clause type identification information (here, clause type ID) that identifies the type of the target clause.

For example, if the classification model stored in the storage 23 does not meet certain performance requirements, the prediction unit 264 uses the converted vector as the vector of each clause contained in the standard contract stored in the storage 23. As a result of comparison, the clause type identification information (here, clause type ID) associated with the storage 23 is output to the vector whose similarity satisfies the standard (here, the most similar vector as an example). As a result, the target clause,
It can be compared with the clauses contained in the standard contract and the type of clause in question can be determined.

More specifically, the prediction unit 264 compares the converted vector with the vector of each clause contained in a standard contract of the same type as the target contract type acquired by the acquisition unit 261. As a result, the target clause can be compared with the clause included in the standard contract of the same type as the target contract type, and the accuracy of determining the type of the target clause can be improved.

Here, at the time of comparison, the prediction unit 264 calculates, for example, the cosine similarity between the converted vector and each of the clauses included in the standard contract, and among the calculated cosine similarity, the vector having the highest cosine similarity. The clause type ID associated with the storage 23 is output to. This makes it possible to grasp the type of clause.

On the other hand, for example, when the classification model stored in the storage 23 satisfies a certain performance requirement, the prediction unit 264 inputs the converted vector into the classification model and identifies the type of the target clause. The identification information (here, clause type ID) is output. Thereby, the accuracy of discriminating the type of clause can be improved.

Then, the prediction unit 264 stores the output clause type ID in the storage 23 in association with the clause number for each clause.

<Conversion process to vector>
Next, an example of vector conversion processing in the conversion unit 263 will be described. FIG. 8 is a schematic diagram showing an example of vectorization of a clause when the number of sentences included in the clause is three.
As shown in FIG. 8, the conversion unit 263 converts the first sentence included in the clause into the vector X1.
Similarly, the conversion unit 263 converts the second sentence included in the clause into the vector X2. Similarly, conversion unit 2
63 converts the third sentence contained in the clause into the vector X3. Then, the conversion unit 263 outputs the average of the vectors (X1 + X2 + X3) / 3 as the vector corresponding to the clause.

Here, an example of the process of converting a sentence into a vector will be described.
(Step 1) First, the conversion unit 263 decomposes the sentence into morphemes.
For example, if the sentence is "Today is good weather", then "Today", "Ha", "Good", "Weather", "
Is divided into.

(Step 2) Next, the conversion unit 263 adds ng of N consecutive morphemes (N is a natural number).
Defined as ram. For example, when N is 2, n-gram is defined as follows.
N = 2: (Today, ha), (ha, good), (good weather), (weather)

(Step 3) If there are three documents S1, S2, and S3, the conversion unit 263 calculates the optimum matrices U and V as follows. That is, all the n-grams of the document S1 appear most in the document S1, and the different n-grams contained in the documents S2 and S3 do not appear. The same applies to the document S2 and the document S3. The matrices U and V are those described in Non-Patent Document 1. The matrix U is a set of values for each sentence (sentence), and the vector (distributed representation) of the document is derived by executing learning (optimization function) by optimizing the matrices U and V. The matrix U is a matrix of the same size as the matrix V and tilted by 90 degrees, and is a set of values for each sentence. Then, the conversion unit 263 obtains a sentence vector (also referred to as a distributed representation) by averaging all the included n-gram distributed representations based on V, which is one of the matrices.
Hereinafter, an example of a specific method of the process of converting the target sentence (sentence) S into a vector will be briefly described. At the time of learning, the expression for each word is obtained by the matrix V. Parameter U is also used to learn the appearance pattern in the sentence of the word. Also, the matrices U and V are optimized using the error function. As the error function, for example, the equation (3) of Non-Patent Document 1 may be used.
At the time of deriving the distributed representation, the Bag of Words vector D of the sentence (sentence) S is obtained. The vector expression of the sentence S is obtained by multiplying the vector D and the matrix V and taking the average of the vector expressions for each n-gram including the frequency of appearance.
A known library or a known method may be used for the process of converting a sentence into a vector.

<About display processing of clause type>
An example of processing for displaying the clause type will be described. FIG. 9 is a first example of a browser screen displayed on the terminal 1. In FIG. 9, when the data of the business transfer contract is read in the terminal 1, it is displayed for each clause in the browser of the terminal 1. In FIG. 9, "Article 1" is a screen when selected by the user. When "Article 1" is selected by the user
As shown in FIG. 7, the clause type requesting unit 164 transmits "1" as the clause number to the server 2 to request the clause type. In response to this request, the clause type reading unit 266 reads the clause type ID stored in association with the clause number "1" from the storage 23. And communication control unit 2
69 is a communication circuit 22 so as to transmit the information corresponding to the read clause type ID to the terminal 1.
To control. When this information is received, the output control unit 166 of the terminal 1 displays a tag corresponding to the clause type ID (here, the ID that identifies "business transfer"). As a result, as shown in FIG. 9, the tag R1 of "business transfer" is attached to "Article 1" and displayed.

<About similar clause output processing>
Next, the similar clause output processing will be described. In FIG. 9, the "search by this tag" button B1 is displayed. Further, the "Delete clause" button B2 is displayed, and the "Add clause" button B3 is displayed. When the "Delete Clause" button B2 is pressed, the processor 16 of the terminal 1 deletes the clause. On the other hand, when the "add clause" button B3 is pressed, the processor 16 of the terminal 1 adds the clause being edited.

Further, in FIG. 9, a select box SB1 for selecting the type of contract and a select box SB2 for selecting the type of clause are displayed. When the contract type is selected in the select box SB1 and the clause type is selected in the select box SB2, the clause data requesting unit 163 of the terminal 1 determines the contract type ID that identifies the selected contract type, and A clause type ID that identifies the selected clause type is transmitted to the server 2 to request clause data of the contract type and corresponding to the clause type. In response to this request, server 2
The clause data reading unit 265 reads the clause associated with the received contract type ID and clause type ID set from the clause table T2 of the storage 23. And communication control unit 1
65 transmits the clause data indicating the read clause to the terminal 1. The output control unit 166 of the terminal 1 displays this clause data on the browser screen. As a result, in the case of FIG. 9, since the business transfer contract is selected as the contract type and the damage compensation is selected as the clause type, the damage compensation clause of the other business transfer contract is displayed in the screen area R2. The contract.

FIG. 10 is an example of a browser screen when the “search by this tag” button B1 is pressed in FIG. In FIG. 9, when the "Search by this tag" button B1 is pressed, the provisions of other contracts of the same type (here, business transfer) to which the same clause type as the clause type assigned to the clause is assigned. Is displayed, for example, in the screen area R3 of FIG. In addition
The value in the select box of the clause tag switches from "damages" to "business transfer".
As a process at that time, when the "Search by this tag" button B1 is pressed, the clause data requesting unit 163 of the terminal 1 identifies the contract type ID and the assigned clause. The clause type ID that identifies the type of the contract is transmitted to the server 2, and the clause data corresponding to the contract type and the clause type is requested. In response to this request, the clause data reading unit 265 of the server 2 reads the clause associated with the received contract type ID and clause type ID pair from the clause table T2 of the storage 23. Then, the communication control unit 165 transmits the clause data indicating the read clause to the terminal 1. Output control unit 16 of terminal 1
6 displays this clause data on the browser screen. As a result, in the case of FIG. 10, in the screen area R3, the business transfer clause of the other business transfer contract is displayed.

FIG. 11 is a flowchart showing an example of processing when the prediction unit assigns a clause type to a clause using the similarity in the present embodiment.
(Step S10) First, the read reception unit 161 of the terminal 1 reads the contract data.

(Step S20) Next, the communication control unit 165 of the terminal 1 transfers this contract data to the server 2.
Send to.

(Step S30) The acquisition unit 261 of the server 2 that has received the contract data acquires the type of the contract from, for example, the title included in the contract.

(Step S40) Next, the division unit 262 of the server 2 divides the contract for each clause.

Hereinafter, the processes of steps S50 to S100 are executed for each of the divided clauses.
(Step S50) The conversion unit 263 of the server 2 translates the terms of the target contract from Japanese to a foreign language (here, English as an example).

(Step S60) Next, the conversion unit 263 of the server 2 converts the translated foreign language (here, English as an example) clause into a vector. This makes it a foreign language (English as an example here)
Since it is converted to a vector after being converted to, it absorbs the notational fluctuations in Japanese, and after converting to a vector, those with similar meanings are converted to vectors with higher similarity, and those with distant meanings are converted to more similarities. Can be converted to a lower vector. Especially by translating into English and vectorizing from English, English is less ambiguous and less syntactically fluctuating than Japanese, so it makes more sense than converting from Japanese to vector when vectorized. It is possible to convert close objects into vectors with higher similarity, and distant objects into vectors with lower similarity.

(Step S70) Next, the prediction unit 264 calculates the similarity (for example, cosine similarity) between the converted vector and each vector of the same type of contract in the standard clause table.

(Step S80) Next, the output unit 267 determines whether or not there is a similarity having a similarity equal to or higher than the setting reference among the calculated similarity.

(Step S90) When there is a similarity whose similarity is equal to or higher than the setting reference among the similarity calculated in step S80, the output unit 267 sets the clause type ID corresponding to the vector having the highest similarity to the clause number. It is saved in the storage 23 in association with.

(Step S100) On the other hand, if there is no similarity calculated in step S80 whose similarity is equal to or higher than the setting reference, the output unit 267 stores the clause number of the clause as a dissimilar clause in the storage 23. To do.

(Step S110) Next, the output unit 267 of the server 2 reads out the clause number of the dissimilar clause.

In this way, the conversion unit 263 converts each of the clauses included in the target contract into a vector. Then, the prediction unit 264 compares each of the converted vectors with the vector of each clause included in the standard contract of the same type as the target contract (here, the cosine similarity is calculated as an example). As a result of comparison by the prediction unit 264, the output unit 267 has a similarity equal to or higher than the setting standard with the vector of any of the clauses included in the standard contract among the converted vectors (for example, the vectors V11 to V16 in FIG. 1). The clause corresponding to the vector without (for example, the two vectors V11 and V16 in FIG. 1) is output as a dissimilar clause. With this configuration, the user can grasp the provisions that are not in the standard contract, so that the contract can be created and / or checked more efficiently.

In step S110, as a result of comparison by the prediction unit 264, the output unit 267 is set between the vector of the clause included in the standard contract (for example, the vectors V21 to V26 in FIG. 1) and the converted vector. The information corresponding to the clause type identification information (here, the clause type ID) indicating the type of clause corresponding to the vector having no similarity above the standard (for example, the two vectors V21 and V26 in FIG. 1) is targeted. It may be output as a clause missing in the contract. With this configuration, the user can grasp the provisions that are not in the standard contract, so that the contract can be created and / or checked more efficiently.

(Step S120) Next, the communication control unit 269 of the server 2 transmits the read clause number of the dissimilar clause and the contract data divided for each clause.

(Step S130) When the output control unit 166 of the terminal 1 acquires the clause number of the similar clause and the contract data divided for each clause, the contract can be displayed separately for each clause and the dissimilar clause can be distinguished. Display on. As a result, the user can grasp the dissimilar clause and assign the clause type to this dissimilar clause.

(Step S140) The processor 16 of the terminal 1 determines whether or not the clause type has been input by the user for the dissimilar clause.

(Step S150) In step S140, when the clause type is input by the user for the similar clause, the clause number of the dissimilar clause and the clause type ID for identifying the clause type entered by the user are transmitted to the server 2. ..

(Step S160) When the server 2 receives the clause number and the clause type ID of the dissimilar clause, the server 2 inputs the vector of the clause specified by the clause number of the received dissimilar clause and outputs the received clause type ID. The dataset is added to the teacher data stored in the storage 23.

(Step S170) The update unit 268 of the server 2 updates the classification model by executing machine learning using the teacher data stored in the storage 23 at regular intervals or at set timings.

In this way, the update unit 268 inputs the vector of the dissimilar clause output by the output unit 267 and outputs the clause type identification information (clause type ID) that identifies the type of the clause assigned by the user to the clause. The classification model is updated by machine learning using the teacher data. Then, the prediction unit 264 inputs the converted vector into the updated classification model, and outputs the clause type identification information (here, the clause type ID) that identifies the type of the target clause. Thereby, by using the updated classification model, it is possible to improve the accuracy of assigning the correct clause type identification information (here, clause type ID) to the clause.

FIG. 12 is a flowchart showing an example of a process of presenting a clause type and a similar clause in the present embodiment. This process is a process executed after the process of FIG. 11 or FIG. 13 described later is completed.

(Step S210) The clause type requesting unit 164 of the terminal 1 determines whether or not the clause has been selected by the user.

(Step S220) When a clause is selected by the user in step S210, the communication control unit 165 controls to transmit the clause number of the clause selected by the user to the server 2.

(Step S230) The clause type reading unit 266 of the server 2 reads the clause type ID corresponding to the clause number from the storage 23.

(Step S240) The communication control unit 269 of the server 2 controls to transmit the information corresponding to the read clause type ID to the terminal 1.

(Step S250) When the output control unit 166 of the terminal 1 receives the information corresponding to this clause type ID, the output control unit 166 browses the information corresponding to this clause type ID (for example, the tag R1 of “business transfer” in FIG. 9). Display more on the screen.

(Step S260) For example, on the browser screen of FIG. 9, the clause data requesting unit 163 of the terminal 1 determines whether or not the "search by this tag" button B1 is pressed.

(Step S270) Step S260 When it is determined that the "Search by this tag" button B1 is pressed, the communication control unit 165 of the terminal 1 sets the clause of the clause to which the "Search by this tag" button B1 is associated. Type ID (In the example of FIG. 9, the clause type ID of "circumstance transfer")
Is controlled to be transmitted to the server 2.

(Step S280) Next, when the clause data reading unit 265 of the server 2 receives the clause type ID, it reads at least one clause associated with this clause type ID as a similar clause in the clause table T2 of the storage 23. .. In this way, the storage 23 stores one or more clause data in association with the clause type identification information (here, clause type ID). When the clause data reading unit (clause data output unit) 265 receives a specific operation of the user for a specific clause included in the target contract, the clause type identification information (here, the clause type identification information) that identifies the type of the specific clause. Same clause type identification information (clause type ID)
Here, the clause data associated with the storage 23 is output to the clause type ID). As a result, it is possible to refer to other clauses of the same type as the clause selected by the user, so that the creation of the contract can be made more efficient.
In addition, in the storage 23, with respect to the clause type identification information (here, clause type ID), in addition to or in place of one or more clause data, change history and risk information (for example, the clause) related to the clause data are stored. Data (information indicating high or low risk), comments, and / or data such as proposed amendments and additional provisions (hereinafter collectively referred to as "clause metadata") are associated and stored. You may. In that case, when the clause data reading unit (clause data output unit) 265 receives a specific operation of the user for a specific clause included in the target contract, the clause type identification information for identifying the type of the specific clause. The clause data and / or clause metadata associated in the storage 23 may be output to the same clause type identification information (here, clause type ID) as (here, clause type ID).

(Step S290) The communication control unit 165 of the server 2 controls to transmit the data of the clause read in step S280 to the terminal 1.

(Step S300) When the output control unit 166 of the terminal 1 receives the clause data, the output control unit 166 further displays the received clause data on, for example, a browser screen. As a result, it is possible to refer to other clauses of the same type as the clause selected by the user, so that the creation of the contract can be made more efficient.

FIG. 13 is a flowchart showing an example of processing when the prediction unit assigns a clause type to a clause using a classification model in the present embodiment. Since the processing of steps S410 to S460 is the same as that of steps S10 to S60 of FIG. 11, the description thereof will be omitted. The processing of steps S450 to S480 is repeated for each clause.

(Step S470) The prediction unit 264 of the server 2 inputs the converted vector into the classification model and outputs a clause type ID that identifies the type of the target clause.

(Step S480) Next, the prediction unit 264 of the server 2 stores the set of the clause number and the clause type ID output in step S470 in the storage 23 for the target clause.

(Step S490) The communication control unit 269 of the server 2 transmits the contract data divided for each clause.

(Step S500) When the output control unit 166 of the terminal 1 receives the contract data divided for each clause, the output control unit 166 displays the contract separately for each clause by using the contract data divided for each clause.

As described above, the information processing system S according to the present embodiment stores the analyzed relationship between the vector in which the clause of the contract is converted and the clause type identification information representing the type of the clause.
3 is provided. Further, the information processing system S includes a conversion unit that converts the target clause of the target contract into a vector. Further, the information processing system S includes a prediction unit 264 that predicts and outputs the clause type identification information that identifies the type of the target clause by inputting the converted vector based on the analyzed relationship.

According to this configuration, by assigning the clause type identification information to the target clause, the user can grasp the type of the target clause, so that the user can grasp what kind of clause the contract clause is. Can be facilitated.

In the present embodiment, the storage 23 stores the analyzed relationship between the vector in which the clause of the contract is converted and the clause type identification information indicating the type of the clause, but the present invention is not limited to this.
In addition to or in lieu of this, the terms of the contract are transformed vectors and clause metadata for the clause (eg, risk of the clause, comments prompting changes or additions to the clause, draft amendments, clauses to be added). Etc.) may be stored in the analyzed relationship. In that case,
Based on this analyzed relationship, the prediction unit 264 may use the converted vector as an input to predict and output the clause metadata of the target clause.

According to this configuration, if the clause metadata suggests which side of the contract is the disadvantageous clause, the target user may change the clause that is disadvantageous to him or her so that it is not disadvantageous to him or her. You can add new clauses so that it does not become. Also, if the clause metadata suggests which side of the contract is the favorable clause, the target user is
You can create and / or check contracts while grasping the clauses that are advantageous to you. This can improve the efficiency of contract creation and / or checking.

Furthermore, in the case of a sales contract, for example, one clause is advantageous or disadvantageous depending on whether the user who is the contracting party is the buyer or the seller (that is, the position of the contracting party), and another clause (that is, the position of the contracting party).
For example, if it is a buyer, whether or not the seller should add a clause for which the seller is liable for defects will change. In addition, the contract is based not only on the position of the contracting party, but also on the classification of the parties such as (1) corporations, individuals, large companies, small and medium-sized enterprises, private companies and other commercial organizations, and universities and other non-profit organizations.
It depends on whether one clause is advantageous or disadvantageous, and whether or not another clause is added. In addition, (2) important index categories, which are important indicators such as the amount of money and the period, should be checked with priority.

The amount of money in (2) is, for example, 10% of the sales amount in the draft of the contract presented by the other party, although it is standard that 5% of the sales amount is the license fee in a certain type of license contract.
If it is stated that% is the license fee, it is disadvantageous for the licensee. In other words, the amount of money stated in the contract also depends on the position of the contracting party or the classification of the parties, whether it is advantageous or disadvantageous.

Therefore, in the information processing system according to the present embodiment, the prediction unit 264 loads the target clause (read by the user) from the viewpoint of the classification of the user who is the contracting party (for example, the position of the contracting party, the classification of the parties, etc.). If the terms of the contract) are unfavorable, you may output clause metadata suggesting changes to the clauses or adding clauses.

Specifically, for example, the vector of the target clause (the clause of the contract read by the user),
The similarity between the target vector and the standard clause vector meets the criteria when compared to the standard clause vector contained in the standard clause data set that can be determined to be advantageous or disadvantageous according to the user's classification and / or important index classification. If so, clause metadata contained in the standard clause dataset (eg, amended clause examples, additional clause examples, comments, etc.) may be output.

<Example 1>
For example, the storage 23 contains contract identification information (for example, a contract ID that identifies a sales contract), a user classification (for example, a buyer), and the user for a certain type of clause (for example, a suspension clause). Standard clauses (for example, X clauses) that are disadvantageous to the division and clause metadata (for example, "This clause is disadvantageous. It is recommended to change it to XX")
Is associated and remembered. In that case, the case where the A clause is included in the clause of the type of the contract provided by the user will be described.

In the processor 26 of the server 2, for example, the A clause is similar to the X clause by the set standard or more (for example, when the cosine similarity between the vector of the A clause and the vector of the X clause is equal to or more than the threshold value) and the target user is the X clause. If it corresponds to the buyer who is the associated user category in the storage 23, the comment "This clause is disadvantageous. It is recommended to change it to XX" may be output as clause metadata.

<Example 2>
For example, the storage 23 contains contract identification information (for example, a contract ID that identifies a sales contract), a user classification (for example, a buyer), and the user for a certain type of clause (for example, a suspension clause). Standard clauses that are detrimental to the category (eg, Y clause) and clause metadata (eg, the comment "We recommend adding the following clause" and the recommended clause C) are associated and stored. ing. In that case, the case where the B clause is included in the clause of the type of the contract provided by the user will be described.

In the prediction unit 264, for example, the B clause is similar to the Y clause by the setting standard or more (for example, when the cosine similarity between the vector of the B clause and the vector of the Y clause is equal to or more than the threshold value) and the target user stores in the Y clause 23. If it corresponds to the buyer who is the user category associated with in, clause C may be output with the comment "It is recommended to add the following clause".
In addition to or in place of the disadvantageous standard clause, the advantageous standard clause may be stored in the storage 23.

As described above, the storage 23 contains the user classification (for example, the buyer), the vector of the standard clause (for example, the Y clause) that is advantageous or disadvantageous for the user classification, and the clause metadata (for example,
For example, the comment "It is recommended to add the following clause" may be stored in association with the recommended clause C). Then, in the prediction unit 264, for example, the vector of the target clause is similar to the vector of the standard clause stored in the storage 23 by the setting standard or more, and the classification of the target user is associated with the vector of the standard clause in the storage 23. If the user category is applicable, the clause metadata associated with the vector of the target clause and the set of the user category in the storage 23 may be output.

With this configuration, if the clause is disadvantageous in the target user's classification, the clause metadata suggesting to change or add the clause is obtained, so the target user changes the clause that is disadvantageous to himself so as not to be disadvantageous. Or you can add new clauses so that you are not at your disadvantage. In addition, in the case of a clause that is advantageous in the classification of the target user, clause metadata suggesting that the clause is advantageous can be obtained, so that the target user can create a contract while grasping the clause that is advantageous to him / herself. / Or can be checked. This can improve the efficiency of contract creation and / or checking.

It should be noted that the storage 23 may store the contract identification information in association with the user classification, the vector of the standard clause advantageous or disadvantageous to the user classification, and the set of clause metadata.

In that case, for example, the prediction unit 264 resembles the vector of the target clause to the vector of the standard clause stored in the storage 23 more than the setting standard, and the contract identification information of the target contract becomes the vector of the standard clause. If the contract identification information associated with the storage 23 matches and the target user category corresponds to the user category associated with the storage 23 vector of the standard clause, the vector of the target clause and the user category The clause metadata associated with the storage 23 may be output to the pair.

In addition to outputting clause metadata based on rules as described above, clause metadata may be output using an advice model obtained by machine learning. Here, the advisory model is, for example, a model machine-learned as teacher data in which a set of a user classification and a vector of standard clauses advantageous or disadvantageous to the user classification is input and clause metadata is output, and is stored in the storage 23. You may be. In this case, the processor 26 of the server 2
The classification of the target user and the vector of the target clause may be input to this advisory model to function as a prediction unit that outputs clause metadata.

As described above, the processor 26 of the server 2 has an analyzed relationship between the user classification, the vector of the standard clause advantageous or disadvantageous to the user classification, and the clause metadata (for example, the above rule-based relation or the above-mentioned rule-based relation or Based on the above-mentioned advice model), it may function as a prediction unit that predicts and outputs clause metadata by inputting the classification of the target user and the vector of the target clause.

According to this configuration, if the clause is disadvantageous in the target user's classification, the clause metadata suggesting to change or add the clause is obtained, so that the target user does not disadvantage the clause that is disadvantageous to himself / herself. You can change it to, or add a new clause so that it does not disadvantage you. In addition, in the case of a clause that is advantageous in the classification of the target user, clause metadata suggesting that the clause is advantageous can be obtained, so that the target user can create a contract while grasping the clause that is advantageous to him / herself. / Or can be checked. This can improve the efficiency of contract creation and / or checking.

Note that this analyzed relationship is not limited to the analysis relationship between the user category, the vector of standard clauses that are advantageous or disadvantageous to the user category, and the clause metadata, but also the contract identification information, the user category, and the user. It may be an analyzed relationship between a set of standard clause vectors that are favorable or unfavorable for the division and clause metadata.

At least a part of the server 2 described in the above-described embodiment may be configured by hardware or software. When configured by hardware, a program that realizes at least a part of the functions of the information processing system 1 may be stored in a recording medium such as a flexible disk or a CD-ROM, read by a computer, and executed. The recording medium is not limited to a removable one such as a magnetic disk or an optical disk, and may be a fixed recording medium such as a hard disk device or a memory.

Further, a program that realizes at least a part of the functions of the server 2 may be distributed via a communication line (including wireless communication) such as the Internet. Further, the program may be encrypted, modulated, compressed, and distributed via a wired line or wireless line such as the Internet, or stored in a recording medium.

Further, the server 2 may be made to function as an information processing system by one or a plurality of information processing devices. When a plurality of information processing devices are used, one of the information processing devices may be a computer, and the computer may execute a predetermined program to realize a function as at least one means of the information processing system. ..

Further, in the invention of the method, all the steps (steps) may be realized by automatic control by a computer. Further, the progress control between the processes may be manually performed while the computer is used to perform each process. Further, at least a part of the whole process may be performed manually.

As described above, the present invention is not limited to the above embodiment as it is, and at the implementation stage, the components can be modified and embodied within a range that does not deviate from the gist thereof. In addition, various inventions can be formed by an appropriate combination of the plurality of components disclosed in the above-described embodiment. For example, some components may be removed from all the components shown in the embodiments. Further, components over different embodiments may be combined as appropriate.

1 Terminal 11 Input interface 12 Communication circuit 13 Storage 14 Memory 15 Output interface 16 Processor 161 Read reception unit 162 Editing reception unit 163 Clause data request section 164 Clause type request section 165 Communication control section 166 Output control section 17 Display 2 Server 21 Input interface 22 Communication circuit 23 Storage 24 Memory 25 Output interface 26 Processor 261 Acquisition unit 262 Division unit 263 Conversion unit 264 Prediction unit 265 Clause data reading section (clause data output section)
266 Clause type Read unit 267 Output unit 268 Update unit 269 Communication control unit S Information processing system

Claims (13)

Storage that stores the analyzed relationship between the transformed vector of the terms of the contract and the clause type identification information representing the type of the clause or the clause metadata for the clause.
A conversion unit that converts the target clause of the target contract into a vector,
Based on the analyzed relationship, a prediction unit that predicts and outputs clause type identification information for identifying the type of the target clause or clause metadata for the target clause by using the converted vector as an input.
Information processing system equipped with. For the analyzed relationship stored in the storage, the vector of the clause that is not similar to the vector of any clause included in the standard contract is input, and the clause type assigned by the user to the clause is input. Identification clause type This is a classification model machine-learned as teacher data that outputs identification information.
The information processing system according to claim 1, wherein the prediction unit inputs the converted vector into the classification model and outputs clause type identification information for identifying the type of the target clause. The analyzed relationship stored in the storage is a set of a vector corresponding to each clause included in the standard contract and a clause type identification information indicating the type of the clause.
The prediction unit compares the converted vector with the vector of each clause included in the standard contract stored in the storage, and as a result of the comparison, associates the vector with a vector whose similarity satisfies the standard in the storage. The information processing system according to claim 1, which outputs the specified clause type identification information. In the storage, the contract type is further associated and stored with respect to the set of the vector corresponding to each clause included in the standard contract and the clause type identification information attached to the clause.
Further equipped with an acquisition unit for acquiring the type of the target contract,
The information according to claim 3, wherein the prediction unit compares the converted vector with the vector of each clause included in a standard contract of the same type as the target contract type acquired by the acquisition unit. Processing system. The conversion unit converts each clause contained in the target contract into a vector.
The prediction unit compares each of the converted vectors with the vector of each clause contained in a standard contract of the same type as the target contract.
As a result of comparison by the prediction unit, the output unit that outputs as a dissimilar clause the clause corresponding to the vector after conversion that does not have a similarity equal to or higher than the setting standard with the vector of any clause included in the standard contract. The information processing system according to claim 3 or 4. By machine learning using the teacher data that inputs the vector of the dissimilar clause output by the output unit and outputs the clause type identification information that identifies the type of clause assigned by the user to the clause. It also has an update section that updates the classification model.
The information processing system according to claim 5, wherein the prediction unit inputs the converted vector into the updated classification model and outputs clause type identification information for identifying the type of the target clause. The conversion unit converts each clause contained in the target contract into a vector.
The prediction unit compares each of the converted vectors with the vector of each clause contained in a standard contract of the same type as the target contract.
As a result of the comparison by the prediction unit, among the vectors of the clauses included in the standard contract, the clause type identification indicating the type of the clause corresponding to the vector having no similarity equal to or higher than the setting standard with the converted vector. The information processing system according to claim 3 or 4, further comprising an output unit that outputs information corresponding to the information as a clause lacking in the target contract. In the storage, one or more clause data and / or clause metadata are associated with the clause type identification information and stored.
When the user accepts a specific operation on a specific clause contained in the target contract, the clause type identification information is the same as the clause type identification information that identifies the type of the specific clause, and the clause associated with the storage. The information processing system according to any one of claims 1 to 7, further comprising a clause data output unit that outputs data and / or clause metadata. The information processing system according to any one of claims 1 to 8, wherein the conversion unit translates the terms of the target contract from Japanese to a foreign language and converts the translated terms of the foreign language into a vector. The information processing system according to claim 9, wherein the foreign language is English. Further provided with a division section for dividing the target contract into clauses,
The information processing system according to any one of claims 1 to 10, wherein the conversion unit converts the clause after division by the division unit into a vector. The storage stores a set of user categories, a vector of standard clauses that are favorable or unfavorable for the user sort, and an analyzed relationship with clause metadata.
The information processing system according to claim 1, wherein the prediction unit predicts and outputs clause metadata by inputting a target user classification and a vector of the target clause based on the analyzed relationship. An information processing method executed by an information processing system that stores storage that stores the analyzed relationship between the vector in which the clause is converted and the clause type identification information that represents the type of the clause or the clause metadata for the clause. There,
Steps to convert the subject clause of the subject contract into a vector,
Based on the analyzed relationship, the step of predicting and outputting the clause type identification information for identifying the type of the target clause or the clause metadata for the target clause by inputting the converted vector.
Information processing method having.