JP6680656B2 - Judgment device and judgment method - Google Patents

Description

  The present invention relates to a determination device and a determination method.

  BACKGROUND ART Conventionally, there is known a technique of searching or generating information related to the input information based on an analysis result of the input information and outputting the searched or generated information as a response. As an example of such a technique, the words, sentences, and contexts included in the input text are converted into a multidimensional vector and analyzed, and based on the analysis result, text similar to the input text or the input text is input. There is a known natural language processing technique in which a text following a text is analogized and the analogy result is output.

JP 2006-127077 A

  However, the above-mentioned conventional technique may not be able to output the information for assisting the user's creation.

  For example, in the above-mentioned conventional technology, only information similar to the input text, text following the input text, or the like that the user can predict is output, and it is related to the input text. However, it is difficult for the user to output information that cannot be predicted. For this reason, the above-mentioned conventional technology cannot provide information that inspires the user.

  The present application has been made in view of the above, and an object thereof is to make it possible to output information that assists the creation of a user.

  The determination device according to the present application, a random matrix indicating a distributed expression space of words belonging to a predetermined learning unit, an acquisition unit that acquires for each learning unit, and a calculation that calculates the eigenvalues of the random matrix acquired by the acquisition unit And a determination unit that determines the similarity of the learning units based on the comparison result of the eigenvalues calculated by the calculation unit.

  According to one aspect of the embodiment, it is possible to enable output of information that assists the creation of the user.

FIG. 1 is a diagram illustrating an example of a determination process executed by the determination device according to the embodiment. FIG. 2 is a diagram illustrating a configuration example of the determination device according to the embodiment. FIG. 3 is a diagram showing an example of information registered in the distributed expression space database according to the embodiment. FIG. 4 is a diagram illustrating an example of information registered in the determination result database according to the embodiment. FIG. 5 is a flowchart illustrating an example of the flow of the determination process according to the embodiment. FIG. 6 is a diagram illustrating an example of the hardware configuration.

  Hereinafter, modes for carrying out the determination device and the determination method according to the present application (hereinafter, referred to as “embodiments”) will be described in detail with reference to the drawings. The determination device and the determination method according to the present application are not limited to this embodiment. Also, in each of the following embodiments, the same parts are designated by the same reference numerals, and duplicated description will be omitted.

[Embodiment]
[1-1. Example of determination device]
First, an example of processing executed by the determination device will be described with reference to FIG. FIG. 1 is a diagram illustrating an example of a determination process executed by the determination device according to the embodiment. In FIG. 1, the determination device 10 is an information processing device that executes determination processing described below, and is realized by, for example, a server device, a cloud system, or the like.

  More specifically, the determination device 10 is an arbitrary device such as the input device 100 or the information processing device 200 (see, for example, FIG. 2) via a predetermined network N such as the Internet (see, for example, FIG. 2). Can communicate with.

  Here, the input device 100 acquires a user's utterance using a voice acquisition device such as a microphone that acquires a voice. Then, the input device 100 converts the utterance into text data using an arbitrary voice recognition technique, and transmits the converted text data to the determination device 10. Further, the information processing device 200 reads out the text data received from the determination device 10 using a device that outputs a voice such as a speaker. The information processing device 200 may display the text data received from the determination device 10 on a predetermined display device.

  The input device 100 and the information processing device 200 are realized by a smart device such as a smartphone and a tablet, an information processing device such as a desktop PC (Personal Computer), a notebook PC, and a server device. The input device 100 and the information processing device 200 may be realized by the same information processing device, for example, a device such as a robot.

[1-2. Regarding the processing of the determination device]
Here, in the prior art, a text similar to the input text or a text following the input text is analogized by using a distributed expression represented by a multidimensional word vector forming the input text. However, in the related art, only text that is similar in distributed expression to the input text, that is, text that the user can predict is output. Therefore, in the conventional technology, it is necessary to provide information that the user does not expect or information that gives a new inspiration to the user, that is, information that may cause the serendipity (inspiration, awareness, surprise) of the user. I couldn't. Also, for example, if a text whose distributed expression is not similar to the input text is output, a text that is not related to the user's thoughts is output, which may disturb the user's thoughts. is there.

  On the other hand, when the user has a concept of a field belonging to a certain field and has a statement having a predetermined conceptual structure, it is a matter that belongs to a field different from the field and the user has a conceptual structure When a sentence having a conceptual structure similar to is output as a response, it may be possible to give a new inspiration to the user. For example, when a user makes a statement that belongs to the field of physics, a response that includes a group of words that belongs to the field of mathematics and that has a similar analogy to the group of words included in the statement of the user. If you do, there is a possibility that you can spread the idea of the user.

  Further, even if the response of the content belonging to the field dissimilar to the field to which the user's statement belongs is output, it may be difficult for the user to understand and the user's thought may be hindered. However, when the response of the content that belongs to the field that is the same as or highly similar to the field to which the user's statement belongs is output, it is considered that only the text of the content that the user can easily imagine can be output. .

  Therefore, the determination device 10 determines the similarity between the fields set with an arbitrary granularity by executing the following determination process. For example, the determination device 10 acquires, for each learning unit, a random matrix indicating a distributed expression space of words belonging to a predetermined learning unit. The determination device 10 also calculates the eigenvalues of the acquired random matrix. Then, the determination device 10 determines the similarity of the learning unit based on the comparison result of the calculated eigenvalues.

  For example, the determination device 10 acquires, as a predetermined learning unit, a random matrix indicating a distributed expression space of words belonging to a predetermined field for each field. Then, the determination device 10 determines the similarity of each field based on the comparison result of the eigenvalues of the random matrix calculated for each field.

  Here, the random matrix is a matrix which is assumed to be given matrix elements according to a predetermined probability law or probability distribution. For example, the determination device 10 converts various words into a distributed expression regardless of the field. For example, the determination device 10 converts a word into a distributed expression using an arbitrary calculation method such as the w2v (word2vec) technique. In the following description, the dispersed expression of words may be described as a word vector.

Subsequently, the determination device 10 specifies a plurality of fields to be compared with arbitrary granularity, and selects, for each specified field, a distributed expression of words belonging to the field. Then, the determination device 10 generates a random matrix A represented by the following Expression (1) from the distributed expression of each word for each field. As a result, the determination device 10 can represent the distributed expression space of words belonging to one field by one random matrix. In the formula (1), vectors of n words belonging to a certain field are described as w _b1 to w _bn, and m predetermined probability distributions are described as w _a1 to w _am . Further, P shown in Expression (1) is a value shown as p in Expression (2) below, and is an expected value of the inner product of two word vectors. That is, the determination device 10 sets a combination matrix of all words in a certain field as a random matrix having expected values as elements.

  Here, when the distributed expression space of words belonging to one field is expressed by a random matrix, it is considered that the characteristics of the random matrix correspond to the general characteristics of words belonging to the field. In addition, it is considered that such overall characteristics of each field serve as an index for determining the similarity between fields.

  Therefore, the determination device 10 regards a random matrix generated from a distributed expression of words belonging to the same field as a random matrix indicating the characteristics of the field (hereinafter, referred to as “random matrix of the field”). , Similarity of each field is determined by comparing the characteristics of the random matrix. More specifically, the determination apparatus 10 calculates the eigenvalue of the random matrix of each field (that is, the characteristic of the random matrix) by using an arbitrary eigenvalue calculation technique such as a known eigenvalue calculation library. Then, the determination device 10 compares the calculated eigenvalues with each other to compare the characteristics of the respective fields, and determines the similarity between the fields based on the comparison result.

  For example, the determination device 10 determines that the similarity between fields is lower than a predetermined threshold when the difference between the eigenvalues is larger than a predetermined threshold, and when the difference between the eigenvalues is equal to or less than the predetermined threshold, It is determined that the similarity between fields is equal to or greater than a predetermined threshold.

  As a result of such processing, the determination device 10 can determine the similarity between the fields based on the comparison result of the overall characteristics of the words belonging to each field. That is, the determination device 10 can determine the similarity between the fields based on the comparison result of the abstracted features of each field. As a result, the determination device 10 selects, for example, another field whose abstract concept is similar to the field to which the text entered by the user belongs, and the text belongs to the selected field and is the text entered by the user. Since a sentence similar in structure to can be output, a sentence that gives the user serendipity can be output as a response.

[1-3. About learning units]
Here, the determination device 10 may set the fields to be compared with arbitrary granularity. For example, when there is a field such as “internal medicine” or “surgery” in the field “medical field”, the determination device 10 determines the eigenvalue of a random matrix generated from a distributed expression of words belonging to the field “internal medicine” and the field “surgery”. The field “internal medicine” and the field “surgery” may be compared by comparing with an eigenvalue of a random matrix generated from the distributed expression of the words belonging to “”. Further, the determination device 10 may compare, for example, the “medical field” including the fields “internal medicine” and the field “surgery” with the “legal field” including the fields “civil law” and the field “criminal law”. .

  Here, in the process of converting a word into a distributed expression such as w2v, a process of converting each word from a sentence or the like belonging to the field into a distributed expression is performed for each predetermined field preset at an arbitrary granularity. Has been done. Therefore, the determination device 10 considers a word belonging to a predetermined learning unit, such as a field set when converting each word into a distributed expression, as a word belonging to a predetermined field, and determines from the distributed expression of the word. The similarity between learning units may be determined by generating a random matrix and comparing the eigenvalues of the random matrix.

[1-4. About the use of eigenvectors]
Here, the features of the random matrix are predicted to appear not only in the eigenvalues but also in the eigenvectors. For example, it is predicted that the eigenvalues of a random matrix of a field may indicate the general concept of words belonging to the field, and the eigenvectors of such a random matrix may indicate the directionality of the general concept of words of the field. It

  Therefore, the determination device 10 may further calculate the eigenvector of the random matrix of each field, and determine the similarity between fields based on the calculated eigenvalue comparison result and the eigenvector comparison result. For example, when the difference between the eigenvalues is equal to or smaller than a predetermined threshold and the value of the cosine similarity of the eigenvector is equal to or larger than the predetermined threshold (when the cosine similarity is equal to or larger than the predetermined threshold), the determination device 10 determines that It may be determined that the fields are similar.

[1-5. About conversion between fields]
Here, based on the eigenvalues of the random matrix, a function (for example, a conversion function) that converts a random matrix of a certain field into a random matrix of another field can be obtained. Such a function can be regarded as a function that changes the concept of words in the field corresponding to the random matrix before conversion into the concept of words in the field corresponding to the random matrix after conversion.

  Therefore, the determination device 10 may generate a conversion function that converts a word belonging to the first learning unit into a word belonging to the second learning unit based on the calculated eigenvalue. Then, the determination device 10 may generate a response corresponding to the sentence input by the user using the generated conversion function.

  For example, the determination device 10 acquires a sentence belonging to the first field from the user. In such a case, the determination device 10 identifies the second field whose similarity to the first field falls within a predetermined range based on the eigenvalues and eigenvectors of the random matrix of the first field. Subsequently, the determination device 10 generates a conversion function that converts the random matrix of the first field into the random matrix of the second field. Note that the determination device 10 does not need to generate a conversion function that completely converts the random matrix of the first field into the random matrix of the second field, and may generate a conversion function that converts with a predetermined accuracy. For example, the determination device 10 may generate a conversion function that matches a predetermined proportion of the matrix elements of the random matrix of the first field with the matrix elements of the random matrix of the second field.

  Then, the determination device 10 extracts a word from the sentence acquired from the user and converts the extracted word into a distributed expression. Then, the determination device 10 uses the conversion function to convert the extracted distributed expression of words into a distributed expression of words that can be regarded as belonging to the second field. Further, the determination device 10 generates a word belonging to the second field from a distributed expression of words that can be regarded as belonging to the second field. For example, the determination apparatus 10 selects a word whose distributed expression is similar to the distributed expression of words that can be regarded as belonging to the second field. Then, the determination device 10 generates a sentence using the selected word and provides the generated sentence to the user as a response. As a result of such processing, the determination device 10 can provide the user with a sentence that expands the scope of ideas and causes serendipity.

[1-6. Example of processing executed by the determination device]
Next, an example of the determination process executed by the determination device 10 will be described with reference to FIG. First, the determination apparatus 10 converts a distributed expression of words belonging to different fields into a random matrix (step S1). For example, the determination apparatus 10 calculates a distributed expression of each of the words # 1-1 to # 1-4 belonging to the field # 1, and uses Equation (1) from the calculated distributed expression to calculate the total of the field # 1. Random matrix # 1 showing specific characteristics is generated. Further, for example, the determination device 10 calculates a distributed expression of each of the words # 2-1 to # 2-4 belonging to the field # 2, and uses Expression (1) from the calculated distributed expression to calculate the field # 2. Generate a random matrix # 2 that shows the overall characteristics of

  Then, the determination apparatus 10 calculates the eigenvalue and eigenvector of each random matrix (step S2). For example, the determination device 10 calculates the eigenvalue # 1 and the eigenvector # 1 from the random matrix # 1, and calculates the eigenvalue # 2 and the eigenvector # 2 from the random matrix # 2.

  Then, the determination device 10 determines the similarity of each field based on the comparison result of the eigenvalues and the comparison result of the eigenvectors (step S3). For example, if the difference between the eigenvalue # 1 and the eigenvalue # 2 is less than or equal to a predetermined threshold and the value of the cosine similarity between the eigenvector # 1 and the eigenvector # 2 is greater than or equal to the predetermined threshold, the determination device 10 determines that It is determined that the field # 1 and the field # 2 are similar fields. Note that the determination device 10 may determine that the field # 1 and the field # 2 are similar when the difference between the eigenvalue # 1 and the eigenvalue # 2 is equal to or less than a predetermined threshold. Further, the determination device 10 comprehensively considers the value of the difference between the eigenvalue # 1 and the eigenvalue # 2 and the value of the cosine similarity between the eigenvector # 1 and the eigenvector # 2, and determines the field # 1 and the field #. It may be judged whether or not 2 is similar. For example, the determination device 10 adopts a threshold value that dynamically changes according to the value of the difference between the eigenvalue # 1 and the eigenvalue # 2, the value of the cosine similarity between the eigenvector # 1 and the eigenvector # 2, and the like. Good. Further, the determination device 10 may set an arbitrary threshold value according to the degree of similarity desired by the user.

  Here, it is assumed that the determination device 10 has acquired the statement of the user (step S4). In such a case, the determination device 10 converts a word included in the sentence of the input statement into a word in a similar field, and generates a sentence having a structure similar to that of the input sentence (step S5). For example, the determination device 10 extracts a word from the input sentence of a statement using a technique such as morphological analysis. Then, the determination device 10 identifies the field to which the user's statement belongs from the extracted words.

  Subsequently, the determination device 10 selects a field similar to the specified field, and uses the conversion function to convert the random matrix of the specified field into the random matrix of the selected field, and includes the sentence in the input statement. Converts the selected words into words in the selected field. For example, when the sentence of the input speech belongs to the field # 1 and the field # 2 exists as a field similar to the field # 1, the determination device 10 sets the random matrix of the field # 1 to the random matrix of the field # 2. The word extracted from the sentence of the input utterance is converted into the word of the field # 2 using the conversion function for converting into the word.

  More specifically, the determination device 10 converts the distributed expression of words extracted from the sentence of the input utterance using a conversion function. Further, the determination device 10 selects a distributed expression similar to the distributed expression converted by using the conversion function from the distributed expressions of the words belonging to the field # 2, and specifies the word corresponding to the selected distributed expression. Then, the determination device 10 uses the selected word to generate a sentence belonging to the field # 2 and having a structure similar to that of the input sentence (that is, a sentence having a similar relationship between words). , The generated sentence is output (step S6). As a result, for example, the determination apparatus 10 can cause the information processing apparatus 200 to speak the generated text as a response.

[2. Configuration of determination device]
Hereinafter, an example of the functional configuration of the determination device 10 that realizes the determination process described above will be described. FIG. 2 is a diagram illustrating a configuration example of the determination device according to the embodiment. As shown in FIG. 2, the determination device 10 includes a communication unit 20, a storage unit 30, and a control unit 40.

  The communication unit 20 is realized by, for example, a NIC (Network Interface Card) or the like. Then, the communication unit 20 is connected to the network N by wire or wirelessly, and transmits and receives information to and from the input device 100 and the information processing device 200.

  The storage unit 30 is realized by, for example, a semiconductor memory device such as a RAM (Random Access Memory) or a flash memory (Flash Memory), or a storage device such as a hard disk or an optical disk. The storage unit 30 also stores a distributed representation space database 31 and a determination result database 32.

  In the distributed expression space database 31, the words converted into the distributed expression are registered for each field. For example, FIG. 3 is a diagram showing an example of information registered in the distributed expression space database according to the embodiment. As shown in FIG. 3, information having items such as “field”, “word”, and “distributed expression” is registered in the distributed expression space database 31.

  Here, the “field” is an identifier indicating the field to which the word indicated by the associated “word” belongs. Further, the "word" is a word corresponding to the distributed expression, that is, a text. The “distributed expression” is a distributed expression of associated “words”. Note that in the example shown in FIG. 3, conceptual values such as “word # 1-1” and “distributed expression # 1-1” are described, but in reality, text data indicating various words and distributed expressions are used. A certain multidimensional quantity or the like is registered.

  For example, in the example shown in FIG. 3, in the distributed expression space database 31, a field “field # 1”, a word “word # 1-1”, and a distributed expression “distributed expression # 1-1” are registered in association with each other. ing. Such information indicates that "word # 1-1" is a word belonging to "field # 1" and its distributed expression is "distributed expression # 1-1".

  Returning to FIG. 2, the description will be continued. In the determination result database 32, the similarity relationship of each field is registered as the determination result of the above-described determination process. For example, FIG. 4 is a diagram illustrating an example of information registered in the determination result database according to the embodiment. As shown in FIG. 4, in the determination result database 32, information having items such as “first field”, “second field” and “conversion function” is registered. Here, the “first field” and the “second field” indicate fields having a similar relationship, and the “conversion function” indicates information indicating a conversion function for converting a random matrix in the field having the associated similar relationship. Is.

  For example, in the example shown in FIG. 4, the first field “field # 1”, the second field “field # 2”, and the conversion function “conversion function # 1” are registered in association with each other. In such information, it is determined that “category # 1” and “category # 2” are similar to each other, and the conversion function for converting the random matrix of “category # 1” into the random matrix of “category # 2” is “ It is a conversion function # 1 ”.

  Returning to FIG. 2, the description will be continued. The control unit 40 is a controller, and various programs stored in a storage device inside the determination device 10 include a RAM and the like by a processor such as a CPU (Central Processing Unit) and an MPU (Micro Processing Unit). It is realized by being executed as a work area. The control unit 40 is a controller, and may be realized by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

  As shown in FIG. 2, the control unit 40 includes an acquisition unit 41, a calculation unit 42, a determination unit 43, a generation unit 44, and an output unit 45.

  The acquisition unit 41 acquires, for each learning unit, a random matrix indicating a distributed expression space of words belonging to a predetermined learning unit. For example, the acquisition unit 41 acquires, as a predetermined learning unit, a random matrix indicating a distributed expression space of words belonging to a predetermined field for each field. To give a more specific example, the acquisition unit 41 refers to the distributed expression space database 31 and selects a distributed expression of words associated with a predetermined field. Then, the acquisition unit 41 uses the selected distributed expression to generate a random matrix in a predetermined field by the above-described equation (1). In addition, the acquisition unit 41 repeats such processing for each field to generate random matrices for all fields.

  The calculation unit 42 calculates each eigenvalue of the random matrix. Further, the calculation unit 42 further calculates the eigenvector of the random matrix. For example, the calculation unit 42 calculates the eigenvalue and the eigenvector from the random matrix generated by the acquisition unit 41 for each field.

  The determination unit 43 determines the similarity of learning units based on the comparison result of the eigenvalues. For example, the determination unit 43 determines the similarity of each field based on the comparison result of the eigenvalues. In addition, the determination unit 43 may determine the similarity of the learning unit based on the comparison result of the eigenvalues and the comparison result of the eigenvectors. For example, the determination unit 43 may determine the similarity of the learning unit based on the comparison result of the eigenvalues and the value of the cosine similarity between the eigenvectors.

  As a more specific example, the determination unit 43 acquires the eigenvalue and the eigenvector calculated by the calculation unit 42 for each field. Then, the determination unit 43 compares the eigenvalues and the eigenvectors with respect to all combinations of fields, and determines whether or not the fields are similar based on the comparison result. For example, the determining unit 43 determines that the difference between the eigenvalue of the field # 1 and the eigenvalue of the field # 2 is less than or equal to a predetermined threshold, and the value of the cosine similarity between the eigenvector of the field # 1 and the eigenvector of the field # 2. When is greater than or equal to a predetermined threshold value, it is determined that the field # 1 and the field # 2 are similar. Then, the determination unit 43 registers the field # 1 and the field # 2 in association with the determination result database 32.

  The generation unit 44 generates a conversion function that converts a word belonging to the first learning unit into a word belonging to the second learning unit based on the eigenvalue. For example, the generation unit 44 refers to the determination result database 32 and identifies the fields # 1 and # 2 determined to be similar by the determination unit 43. In such a case, the generation unit 44 calculates the conversion function # 1 for converting the random matrix of the field # 1 into the random matrix of the field # 2 from the eigenvalue of the field # 1 and the eigenvalue of the field # 2. Then, the generation unit 44 registers the calculated conversion function # 1 in the determination result database 32 in association with the field # 1 and the field # 2. The generation unit 44 may calculate the conversion function using the eigenvector.

  Further, the generation unit 44 uses the conversion function to generate a sentence belonging to the second learning unit from the sentence belonging to the first learning unit input by the user, as a response to the sentence input by the user. To do. For example, when the generating unit 44 receives a sentence of the user's statement from the input device 100, the generating unit 44 extracts a word from the sentence and specifies the field to which the extracted word belongs by using the distributed expression space database 31 or the like. Subsequently, the generation unit 44 refers to the determination result database 32 and selects from the determination result database 32 a conversion function that converts the random matrix of the specified field into a random matrix of a field similar to the specified field.

  Then, the generation unit 44 uses the conversion function selected from the determination result database 32 to convert the word extracted from the sentence of the user's statement into a word in a similar field. After that, the generation unit 44 generates a sentence having the same structure as the sentence input by the user, using the converted word.

  The output unit 45 outputs the sentence generated by the generating unit 44 as a response to the sentence input by the user. For example, the output unit 45 outputs the sentence generated by the generating unit 44 to the information processing apparatus 200, and instructs the information processing apparatus 200 to output the outputted sentence as a response to the utterance received from the input device 100. As a result, the information processing device 200 can output a response.

[3. Example of flow of processing executed by determination device]
Next, an example of the flow of the determination process executed by the determination device 10 will be described with reference to FIG. FIG. 5 is a flowchart illustrating an example of the flow of the determination process according to the embodiment. First, the determination apparatus 10 extracts a word group belonging to each field (step S101), and acquires a random matrix indicating a distributed expression space including a distributed expression of word groups belonging to the same field (step S102). That is, the determination device 10 acquires a random matrix of each field. Then, the determination device 10 calculates the eigenvalues and eigenvectors of the acquired random matrix (step S103), and determines the similarity between fields based on the comparison result of the calculated eigenvalues and eigenvectors (step S104), The process ends.

[4. Modification)
In the above, an example of the determination process by the determination device 10 has been described. However, the embodiment is not limited to this. Hereinafter, variations of the determination process executed by the determination device 10 will be described.

[4-1. Regarding the use of judgment results]
For example, the determination device 10 may execute the process of determining the similarity of the models by using the determination process described above. For example, the determination device 10 generates a first random matrix from a distributed expression group generated from a certain word group using the first model, and generates a first random matrix from the same word group using the second model. Generate a second random matrix from the group. Then, the determination device 10 may determine the similarity between the first model and the second model based on the result of comparison between the eigenvalues of the first random matrix and the eigenvalues of the second random matrix. When such a process is executed, for example, the determination device 10 can determine whether or not the second model can be used instead of the first model.

[4-2. Device configuration〕
In the example described above, the determination device 10 executes the determination process in the determination device 10. However, the embodiment is not limited to this. For example, the determination device 10 accepts from a user using a back-end server that determines the similarity between fields based on the comparison result of eigenvalues and eigenvectors of a random matrix and the determination result of the similarity and the conversion function. It may be realized by a front-end server that converts a sentence and outputs it as a response. Further, the determination device 10 may store the distributed representation space database 31 and the determination result database 32 in an external storage server.

[4-3. Other]
Further, of the processes described in the above embodiment, all or part of the processes described as being automatically performed may be manually performed, or the processes described as manually performed may be performed. All or part of the process can be automatically performed by a known method. In addition, the processing procedures, specific names, and information including various data and parameters shown in the above text and drawings can be arbitrarily changed unless otherwise specified. For example, the various information shown in each drawing is not limited to the information shown.

  Further, each component of each device shown in the drawings is functionally conceptual, and does not necessarily have to be physically configured as shown. That is, the specific form of distribution / integration of each device is not limited to that shown in the figure, and all or a part of the device may be functionally or physically distributed / arranged in arbitrary units according to various loads and usage conditions. It can be integrated and configured.

  Further, the respective embodiments described above can be appropriately combined within a range in which the processing content is not inconsistent.

[5. program〕
Further, the determination device 10 according to the above-described embodiment is realized by, for example, a computer 1000 having a configuration shown in FIG. FIG. 6 is a diagram illustrating an example of the hardware configuration. The computer 1000 is connected to an output device 1010 and an input device 1020, and an arithmetic device 1030, a primary storage device 1040, a secondary storage device 1050, an output IF (Interface) 1060, an input IF 1070, and a network IF 1080 are connected by a bus 1090. Have.

  The arithmetic unit 1030 operates based on a program stored in the primary storage device 1040 or the secondary storage device 1050, a program read from the input device 1020, or the like, and executes various processes. The primary storage device 1040 is a memory device such as a RAM that temporarily stores data used by the arithmetic device 1030 for various calculations. The secondary storage device 1050 is a storage device in which data used by the arithmetic device 1030 for various calculations and various databases are registered, and is realized by a ROM (Read Only Memory), an HDD, a flash memory, or the like.

  The output IF 1060 is an interface for transmitting information to be output to an output device 1010 that outputs various kinds of information such as a monitor and a printer. For example, a USB (Universal Serial Bus) or a DVI (Digital Visual Interface), It is realized by a connector of a standard such as HDMI (registered trademark) (High Definition Multimedia Interface). The input IF 1070 is an interface for receiving information from various input devices 1020 such as a mouse, a keyboard, and a scanner, and is realized by, for example, USB.

  The input device 1020 is, for example, an optical recording medium such as a CD (Compact Disc), a DVD (Digital Versatile Disc), and a PD (Phase change rewritable Disk), a magneto-optical recording medium such as an MO (Magneto-Optical disk), and a tape. It may be a device that reads information from a medium, a magnetic recording medium, a semiconductor memory, or the like. The input device 1020 may also be an external storage medium such as a USB memory.

  The network IF 1080 receives data from another device via the network N and sends the data to the arithmetic device 1030, and also transmits the data generated by the arithmetic device 1030 via the network N to another device.

  The arithmetic device 1030 controls the output device 1010 and the input device 1020 via the output IF 1060 and the input IF 1070. For example, the arithmetic device 1030 loads a program from the input device 1020 or the secondary storage device 1050 onto the primary storage device 1040, and executes the loaded program.

  For example, when the computer 1000 functions as the determination device 10, the arithmetic device 1030 of the computer 1000 implements the function of the control unit 40 by executing the program loaded on the primary storage device 1040.

[6. effect〕
As described above, the determination device 10 acquires, for each learning unit, a random matrix indicating the distributed expression space of words belonging to a predetermined learning unit. The determination device 10 also calculates the eigenvalues of the acquired random matrix. Then, the determination device 10 determines the similarity of the learning unit based on the comparison result of the calculated eigenvalues. As a result, the determination device 10 can specify, for example, a field in which the relative concept is similar, so that the sentence that is similar in structure to the sentence input by the user and is not the sentence input by the user. You can output sentences in different fields. As a result, the determination device 10 can output information that assists the creation of the user.

  Further, the determination device 10 acquires, as a predetermined learning unit, a random matrix indicating a distributed expression space of words belonging to a predetermined field for each field. Then, the determination device 10 determines the similarity of the fields based on the comparison result of the eigenvalues. As a result, the determination device 10 can determine the similarity of the fields based on the comparison result of the overall concept of each field.

  Further, the determination device 10 further calculates the eigenvectors of the random matrix, and determines the similarity of the learning unit based on the comparison result of the eigenvalues and the comparison result of the eigenvectors. For example, the determination device 10 determines the similarity of the learning unit based on the comparison result of the eigenvalues and the value of the cosine similarity between the eigenvectors. As a result, the determination device 10 can improve the determination accuracy of the similarity of fields.

  In addition, the determination device 10 generates a conversion function that converts a word belonging to the first learning unit into a word belonging to the second learning unit based on the calculated eigenvalue. Further, the determination device 10 uses a conversion function to generate a sentence belonging to the second learning unit from a sentence belonging to the first learning unit input by the user, as a response to the sentence input by the user. To do. For this reason, the determination device 10 recognizes that a sentence that the user has not imagined from a sentence that the user has input is a sentence that the user has input in a field different from the field to which the user-input text belongs. It is possible to output a sentence having a similar context. As a result, the determination device 10 can output information that assists the creation of the user.

  As described above, some of the embodiments of the present application have been described in detail based on the drawings, but these are examples, and various modifications based on the knowledge of those skilled in the art, including the modes described in the section of the disclosure of the invention, It is possible to implement the present invention in other forms with improvements.

  Further, the above-mentioned "section (module, unit)" can be read as "means" or "circuit". For example, the generation unit can be read as a generation unit or a generation circuit.

10 determination device 20 communication unit 30 storage unit 31 distributed representation space data database 32 determination result database 40 control unit 41 acquisition unit 42 calculation unit 43 determination unit 44 generation unit 45 output unit 100 input device 200 information processing device

Claims (7)

An acquisition unit that acquires, for each learning unit, a random matrix indicating a distributed expression space of words belonging to a predetermined learning unit,
A calculation unit that calculates the eigenvalues of the random matrix acquired by the acquisition unit,
A determination unit that determines the similarity of the learning units based on the comparison result of the eigenvalues calculated by the calculation unit. The acquisition unit acquires, as the predetermined learning unit, a random matrix indicating a distributed expression space of words belonging to a predetermined field for each field,
The determination device according to claim 1, wherein the determination unit determines similarity of the fields based on a comparison result of the eigenvalues. The calculation unit further calculates an eigenvector of the random matrix,
The determination device according to claim 1 or 2, wherein the determination unit determines the similarity of the learning units based on a comparison result of the eigenvalues and a comparison result of the eigenvectors. The determination device according to claim 3, wherein the determination unit determines the similarity of the learning unit based on a comparison result of the eigenvalues and a value of a cosine similarity between the eigenvectors. The method further comprising: a generation unit that generates a conversion function that converts a word belonging to the first learning unit into a word belonging to the second learning unit based on the eigenvalue calculated by the calculation unit. The determination device according to any one of 4 to 4. The generation unit uses the conversion function to generate a sentence belonging to the second learning unit from a sentence belonging to the first learning unit input by the user, as a response to the sentence input by the user. The determination device according to claim 5, wherein: A determination method performed by the determination device,
An acquisition step of acquiring, for each learning unit, a random matrix indicating a distributed expression space of words belonging to a predetermined learning unit;
A calculation step for calculating the eigenvalues of the random matrix acquired by the acquisition step,
A determination step of determining the similarity of the learning units based on the comparison result of the eigenvalues calculated by the calculation step.

Images