JP7191763B2 - Conversion device, conversion method and conversion program - Google Patents

Description

The present invention relates to a conversion device, a conversion method and a conversion program.

In recent years, techniques for performing language processing using language models have been known. As an example of such technology, there is known a technology for selecting the optimal language model for a user's uttered text from among a plurality of language models.

JP 2018-173846 A JP 2018-55670 A

However, with the above-described conventional technology, it is not always possible to provide content with appropriate expressions to users.

For example, in the above-described conventional technology, language processing that applies a language model based on accompanying attributes associated with the user's user attributes to the uttered text from the user, and when generating a similar sentence from the original text , it merely distinguishes between accepted sentences that can be adopted as similar sentences and rejected sentences that cannot be adopted. Therefore, with the above-described conventional technology, it is not always possible to provide the user with appropriate content.

The present application has been made in view of the above, and aims to provide users with content with appropriate expressions.

A conversion device according to the present application includes a specifying unit that specifies related content that is related to a user to whom the provided content is provided, and an expression of the provided content based on the expression of the related content that is specified by the specifying unit. and a conversion unit that converts the

According to one aspect of the embodiment, there is an effect that it is possible to provide the user with content with appropriate expressions.

FIG. 1 is a diagram illustrating an example of conversion processing according to an embodiment. FIG. 2 is a diagram illustrating an example of integrated language model learning processing according to the embodiment. FIG. 3 is a diagram illustrating a configuration example of a conversion system according to the embodiment; FIG. 4 is a diagram illustrating a configuration example of a conversion device according to the embodiment; 5 is a diagram illustrating an example of a related content storage unit according to the embodiment; FIG. FIG. 6 is a flowchart illustrating an example of a procedure of conversion processing according to the embodiment. FIG. 7 is a hardware configuration diagram showing an example of a computer that implements the functions of the conversion device.

Embodiments for implementing the conversion device, conversion method, and conversion program according to the present application (hereinafter referred to as "embodiments") will be described in detail below with reference to the drawings. Note that the conversion device, conversion method, and conversion program according to the present application are not limited to this embodiment. Also, in each of the following embodiments, the same parts are denoted by the same reference numerals, and overlapping descriptions are omitted.

[1. embodiment]
The conversion processing realized by the conversion device and the like of this embodiment will be described with reference to FIG. FIG. 1 is a diagram illustrating an example of conversion processing according to an embodiment. In the example shown in FIG. 1, the conversion system 1 has a terminal device 10, a viewing content server 20, an input content server 30, and a conversion device 100. FIG. The terminal device 10, the viewing content server 20, the input content server 30, and the conversion device 100 are communicably connected by wire or wirelessly via a predetermined communication network (not shown). Note that the conversion system 1 shown in FIG. 1 may include a plurality of terminal devices 10, a plurality of viewing content servers 20, a plurality of input content servers 30, and a plurality of conversion devices 100. FIG.

A terminal device 10 shown in FIG. 1 is an information processing device used by a user. For example, the terminal device 10 is a desktop PC (Personal Computer), a notebook PC, a tablet terminal, a mobile phone, a PDA (Personal Digital Assistant), or the like. In the example shown in FIG. 1, the terminal device 10 is a smart device such as a smart phone or a tablet used by the user. In the example shown in FIG. 1, the terminal device 10 has an output unit such as a liquid crystal display (hereinafter sometimes referred to as a screen) and adopts a touch panel. That is, the user performs various operations of the terminal device 10 by touching the display surface of the output unit with a finger or a dedicated pen.

The viewing content server 20 shown in FIG. 1 is a server device that provides the terminal device 10 with viewing content viewed by the user. The viewing content server 20 also manages the viewing history of viewing content viewed by the user.

The input content server 30 shown in FIG. 1 is a server device that provides various contents to the terminal device 10 and receives input contents related to character strings input (posted) by users in various contents. The input content server 30 also manages posted information about input content input by the user.

The conversion device 100 shown in FIG. 1 is a server device that provides various contents to the terminal device 10 (hereinafter, content provided to the terminal device 10 by the conversion device 100 may be referred to as "provided content"). In the example of FIG. 1, the conversion device 100 acquires a viewing history of viewing content viewed by the user and a posting history of input content input by the user, and acquires information related to related content related to the user. , and stored in the related content storage unit. Then, the conversion device 100 converts the expression of the content to be provided using a model generated based on the information stored in the related content storage unit, and provides the content to the terminal device 10 . In the example of FIG. 1, the conversion device 100 stores thesaurus information in which concepts having a hierarchical structure are systematically summarized in the storage unit.

The content provided to the terminal device 10 by the viewing content server 20, the input content server 30, and the conversion device 100 is, for example, a portal site, a news site, an auction site, a weather forecast site, a shopping (electronic commerce) site, a finance (stock price ) site, route search site, map providing site, travel site, restaurant introduction site, SNS (Social Networking Service) site, web blog, and the like. The content provided by the viewing content server 20, the input content server 30, and the conversion device 100 is, for example, content related to various applications installed in the terminal device 10 (hereinafter sometimes referred to as "apps"). good too. Specifically, portal apps, news apps, auction sites, weather forecast apps, shopping apps, finance (stock prices) apps, route search apps, map providing apps, travel apps, restaurant introduction apps, SNS apps, blog browsing apps, etc. may be content related to

The conversion processing performed by the conversion device 100 will be described below with reference to FIG. Note that the following description shows an example in which the terminal device 10 is used by the user U1.

First, the terminal device 10 displays viewing content (step S11). For example, in the example of FIG. 1, the terminal device 10 acquires the viewing content from the viewing content server 20 according to the operation of the user U1, and displays it on the screen. Then, the terminal device 10 transmits to the viewing content server 20 the manner in which the viewing content is viewed by the user U1 (hereinafter sometimes referred to as “viewing manner”). For example, in the example of FIG. 1, the terminal device 10 transmits information about the scroll speed of the user U1 with respect to the viewing content displayed on the screen to the viewing content server 20 as the viewing mode. Then, the viewing content server 20 stores information associated with the viewing content corresponding to the viewing mode as a viewing history in the storage unit.

Subsequently, the terminal device 10 transmits the input content input by the user U1 to the input content server 30 among the content provided from the input content server 30 (step S12). For example, in the example of FIG. 1, the terminal device 10 can be used for search queries input by the user U1, reviews input by the user U1 on shopping sites, restaurant introduction sites, and the like, news sites, SNS sites, and the like. transmits the input comment to the input content server 30 as input content. In addition, the terminal device 10 transmits to the input content server 30 the manner in which the user U1 has input the input content (hereinafter sometimes referred to as “input manner”). For example, in the example of FIG. 1, the terminal device 10 transmits to the input content server 30 information about the speed at which the user U1 inputs input content (input speed) as an input mode. Then, the input content server 30 stores information associated with the input content corresponding to the input mode as a posting history in the storage unit.

Subsequently, the conversion device 100 collects viewing histories from the viewing content server 20, and identifies related content related to the user U1 (step S13). For example, in the example of FIG. 1, the conversion device 100 identifies the viewing content viewed by the user U1 as the related content, and stores it in the related content storage unit together with the corresponding viewing mode.

Subsequently, the conversion device 100 collects posting histories from the input content server 30, and identifies related content related to the user U1 (step S14). For example, in the example of FIG. 1, the conversion device 100 identifies the input content input by the user U1 as the related content, and stores it in the related content storage unit together with the corresponding input mode.

Subsequently, the conversion device 100 extracts the easy part and the difficult part of the related content, and generates an easy language model trained on the easy part and a difficult language model trained on the difficult part (step S15). ). Here, the part of the viewing content that the user scrolls at a high speed is the part that does not require time for the user to understand (for example, the part that is expressed in words or writing styles that the user encounters on a daily basis). ), which is assumed to be easy for users to view. On the other hand, the part of the viewing content that the user scrolls slowly is the part that takes time for the user to understand (for example, the part expressed with unfamiliar words or writing style). It is assumed that viewing is a non-easy part. Therefore, in the example of FIG. 1, when the related content is viewing content, the conversion device 100 selects a portion of the related content for which the user U1's scrolling speed is equal to or higher than a predetermined threshold so that the user U1 can easily view the portion of the related content. is extracted as an easy part estimated to be Then, the conversion device 100 extracts a portion where the user U1's scrolling speed is equal to or less than a predetermined threshold as a difficult portion that is estimated to be difficult for the user U1 to view.

In addition, it is assumed that the part of the input content that the user inputs at a high speed is the part that is expressed by the words and writing styles that the user encounters on a daily basis, and that it is easy for the user to view. be. On the other hand, portions of the input content for which the user's input speed is slow are portions expressed by unfamiliar words, style of writing, etc., and are assumed to be portions that are difficult for the user to view. Therefore, in the example of FIG. 1, when the related content is the input content, the conversion device 100 converts the portion of the related content where the user U1's input speed is greater than or equal to a predetermined threshold value to a portion that the user U1 can easily input. (In other words, it is easy for the user U1 to view) is extracted as an easy portion. Then, the conversion device 100 converts the part where the input speed of the user U1 is equal to or less than a predetermined threshold value to a non-easy input that is estimated to be difficult for the user U1 (in other words, difficult for the user U1 to view). Extract as part.

Then, the conversion device 100 generates an easy language model and a non-easy language model based on the character strings included in the extracted easy part and non-easy part. For example, in the example of FIG. 1, the conversion device 100 converts words and sentences extracted from easy parts by techniques such as morphological analysis into vectors (multidimensional quantities) using w2v (word2vec), s2v (sentence2vec), etc. An easy language model trained based on the vectors is generated. The conversion device 100 also converts the words and sentences extracted from the difficult part into vectors, and generates a difficult language model trained based on the vectors.

In the example of FIG. 1, the easy language model is learned so that when a word or sentence is input, a score indicating the degree of similarity to the word or sentence group extracted from the easy part of the related content is output. It is assumed that For example, in the example of FIG. 1, the easy language model converts the input words and sentences into vectors, and the Euclidean distance on the vector space between the vectors corresponding to the words and sentences extracted from the easy part of the related content. Output the score accordingly.

In addition, in the example of FIG. 1, the difficult language model is estimated to be difficult for the user U1 to view when provided content provided by the conversion device 100 to the user U1 is input. It is assumed that learning is performed so as to output the difficult part. For example, in the example of FIG. 1, the difficult language model outputs portions of the provided content that are similar to the difficult portions of the related content as difficult portions. Specifically, the difficult language model converts the words and sentences extracted from the provided content into vectors using techniques such as morphological analysis, and compares them with the vectors corresponding to the words and sentences extracted from the difficult part of the related content. Provided Content output as the non-easy part of

For learning the easy language model and the hard language model, for example, known techniques using various neural networks such as RNN (Recurrent Neural Network) and LSTM (Long short-term memory) can be adopted.

Subsequently, the conversion device 100 receives a request for providing content to be provided from the terminal device 10 (step S16). Subsequently, the conversion device 100 uses the difficult language model to extract a difficult part from the provided content corresponding to the provision request (step S17). For example, in the example of FIG. 1, the conversion device 100 inputs words and sentences extracted from the provided content by techniques such as morphological analysis into the difficult language model, and presumes that viewing of the provided content is difficult for the user U1. Extract the non-easy parts that are

Subsequently, the conversion device 100 uses the thesaurus and the easy language model to convert expressions included in the difficult part of the provided content (step S18). For example, in the example of FIG. 1, the conversion device 100 uses a thesaurus to extract, as candidates, synonyms and similar words for words and sentences extracted from the difficult part of the provided content. Then, the conversion device 100 converts the expression of the difficult part of the provided content using the candidate with the highest score output by the easy language model (“candidate #2”) among the extracted candidates.

Subsequently, the conversion device 100 provides the converted provided content to the terminal device 10 (step S19).

Although the example of FIG. 1 shows an example in which an easy language model and a hard language model are generated and used for conversion processing, models used for conversion processing are not limited to such examples. For example, the conversion device 100 may generate an integrated language model that integrates the easy language model and the difficult language model, and perform learning. Here, the integrated language model learning process will be described with reference to FIG. FIG. 2 is a diagram illustrating an example of integrated language model learning processing according to the embodiment.

As shown in FIG. 2, the conversion device 100 uses an integrated language model that outputs synonyms and similar words when words and sentences are input, and expresses the input words and sentences (hereinafter referred to as “input expressions”). The integrated language model is trained so that the intention of the output synonyms and similar words (hereinafter sometimes referred to as "output expression") is similar to the intention of the output. For example, the conversion device 100 uses w2v, s2v, or the like to determine whether the Euclidean distance between a vector obtained by converting an input word or sentence and a vector obtained by converting an output synonym or similar word is within a predetermined threshold in the vector space. learn to become

In addition, in the example of FIG. 2, the conversion device 100 improves the similarity of the expression of the easy part extracted from the related content (hereinafter sometimes referred to as "easy expression") and the output expression, The integrated language model is trained so that the similarity between the expression of the difficult part (hereinafter sometimes referred to as "difficult expression") and the output expression is reduced. For example, the conversion device 100 uses w2v, s2v, or the like to convert a vector obtained by converting an input word or sentence, and a vector obtained by converting a character string included in an easy part or a non-easy part according to the Euclidean distance on the vector space. The integrated language model is trained based on the similarity.

In this manner, the conversion device 100 according to the embodiment converts the representation of content to be provided to the user based on the content usage history of the user.

Since the difficulty users feel with words varies greatly depending on the generation and experience of the users, it is important to simplify the contents according to the users. Conventional technology uses a method that mechanically converts words defined as difficult on a specific axis into terms that are defined as easy, but this method does not take into account differences in generations and experience, and is It's not always easy.

On the other hand, the conversion device 100 according to the embodiment converts a portion of the content to be provided to the user, which is estimated to be difficult for the user to view, into a simple expression based on the content usage history of the user. Convert. In other words, the conversion device 100 according to the embodiment can provide the user with content with appropriate expressions.

[2. Configuration of conversion system]
Next, the conversion system 1 for realizing the display processing described above will be described with reference to FIG. FIG. 3 is a diagram illustrating a configuration example of a conversion system according to the embodiment; As shown in FIG. 3 , the conversion system 1 includes a terminal device 10 , a viewing content server 20 , an input content server 30 and a conversion device 100 . The terminal device 10, the viewing content server 20, the input content server 30, and the conversion device 100 are connected via a network N so as to be communicable by wire or wirelessly. The network N is a communication network such as a LAN (Local Area Network), a WAN (Wide Area Network), a telephone network (mobile telephone network, fixed telephone network, etc.), a local IP (Internet Protocol) network, or the Internet. The network N may include wired networks or wireless networks. Further, the conversion system 1 shown in FIG. 2 may include a plurality of terminal devices 10, a plurality of viewing content servers 20, a plurality of input content servers 30, and a plurality of conversion devices 100. FIG.

The terminal device 10 is an information processing device used by a user. The terminal device 10 displays information provided by the viewing content server 20, the input content server 30, and the conversion device 100 using a web browser or an application. Further, the terminal device 10 receives control information for realizing display processing of information provided by the viewing content server 20, the input content server 30 and the conversion device 100 from the viewing content server 20, the input content server 30 and the conversion device 100. In this case, the display processing is implemented according to the control information. In addition, the terminal device 10 transmits to the viewing content server 20 the mode (viewing mode) in which the user views the viewing content. In addition, it transmits to the input content server 30 the input content input by the user in the content provided from the input content server 30 and the mode (input mode) when the user inputs the input content.

Note that the terminal device 10 may transmit the viewing content viewed by the user and the viewing mode information to the conversion device 100 . In addition, the terminal device 10 may transmit input content input by the user and information on the input mode to the conversion device 100 .

The viewing content server 20 is a server device that provides the terminal device 10 with viewing content viewed by the user. In addition, the viewing content server 20 stores, in the storage unit, information associating the viewing mode transmitted from the terminal device 10 with the corresponding viewing content as a viewing history. Note that the viewing content server 20 may be a server that distributes application data itself regarding viewing content. Also, the viewing content server 20 may function as a distribution device that distributes control information to the terminal device 10 . Here, the control information is described in, for example, a script language such as JavaScript (registered trademark) or a style sheet language such as CSS (Cascading Style Sheets). Note that the application itself related to the viewing content distributed from the viewing content server 20 may be regarded as the control information.

The input content server 30 is a server device that provides various contents to the terminal device 10 and receives input contents related to character strings input by the user in the various contents. In addition, the input content server 30 stores, in the storage unit, information associating the input mode transmitted from the terminal device 10 with the corresponding input content as a posting history. Note that the input content server 30 may be a server that distributes application data itself regarding content. Also, the input content server 30 may function as a distribution device that distributes control information to the terminal device 10 . Here, the control information is described in, for example, a script language such as JavaScript (registered trademark) or a style sheet language such as CSS. Note that the application itself related to the content distributed from the input content server 30 may be regarded as the control information.

The conversion device 100 is a server device that provides provided content to the terminal device 10 . The conversion device 100 acquires a viewing history and a posting history, and stores them in the related content storage unit 121, which will be described later, as information related to related content related to the user. Then, the conversion device 100 converts the expression of the provided content using a model generated based on the information stored in the related content storage unit 121 and provides the terminal device 10 with the converted content. In addition, the conversion device 100 stores thesaurus information that systematically organizes concepts having a hierarchical structure in the storage unit. Note that the conversion device 100 may be a server that distributes the application data itself regarding the provided content. Also, the conversion device 100 may function as a distribution device that distributes control information to the terminal device 10 . Here, the control information is described in, for example, a script language such as JavaScript (registered trademark) or a style sheet language such as CSS. Note that the application itself related to the content distributed from the conversion device 100 may be regarded as the control information. The configuration of the conversion device 100 will be described later.

[3. Configuration of conversion device]
Next, the configuration of the conversion device 100 will be described with reference to FIG. FIG. 4 is a diagram illustrating a configuration example of a conversion device according to the embodiment; As shown in FIG. 4, the conversion device 100 has a communication section 110, a storage section 120, and a control section .

(Regarding communication unit 110)
The communication unit 110 is realized by, for example, a NIC (Network Interface Card) or the like. The communication unit 110 is connected to the network N by wire or wirelessly, and transmits and receives information to and from the terminal device 10, the viewing content server 20, the input content server 30, and the like.

(Regarding storage unit 120)
The storage unit 120 is realized by, for example, a semiconductor memory device such as a RAM (Random Access Memory) or a flash memory, or a storage device such as a hard disk or an optical disk. As shown in FIG. 4 , storage unit 120 has related content storage unit 121 and model storage unit 122 .

(Regarding Related Content Storage Unit 121)
The related content storage unit 121 stores information related to related content related to the user. An example of information stored in the related content storage unit 121 will now be described with reference to FIG. 5 is a diagram illustrating an example of a related content storage unit according to the embodiment; FIG. In the example of FIG. 5, the related content storage unit 121 has items such as "user ID", "content information", and "mode".

"User ID" indicates identification information for identifying a user. "Content information" indicates information about related content. Note that FIG. 5 shows an example in which conceptual information such as "content information #1" is stored in "content information". Text information and the like corresponding to viewing contents and input contents are stored.

The "mode" is information about the viewing mode of the user's viewing content and the input mode of the user's input content, and has items such as "behavior details" and "operation details". The 'activity content' indicates information indicating the behavior of the user with respect to the content, and stores information such as 'viewing' indicating that the viewing content has been viewed, and 'input' indicating that the input content has been input. The “operation content” indicates the operation content (viewing mode) when the user views the viewing content or the operation content (input mode) when inputting the input content. Note that FIG. 5 shows an example in which conceptual information such as "operation content #1" is stored in "operation content", but in reality, a predetermined range (for example, words, paragraphs, sentences, etc.) of viewing content is stored. , a range defined based on phrases, etc.), an input speed for each predetermined range of input content, and other information is stored.

That is, in FIG. 5, the user identified by the user ID "U1" views the content corresponding to "content information #1", and the operation details at the time of viewing is "operation details #1". indicates

(Regarding the model storage unit 122)
The model storage unit 122 stores an expression model that has learned the characteristics of the expression of the related content. For example, in the example of FIG. 1, the model storage unit 122 stores an easy language model, a difficult language model, and an integrated language model that integrates the easy language model and the difficult language model. Note that the model storage unit 122 may store an expression model generated for each user. In addition, the model storage unit 122 may classify users based on their psychographic attributes, demographic attributes, viewing history, posting history, search history, etc., and store expression models associated with each classification. . For example, the model storage unit 122 stores expression models that have learned related content corresponding to users belonging to the same classification, in association with the classification.

Note that the model storage unit 122 may store information about the generated thesaurus by analyzing the related content using techniques such as morphological analysis.

(Regarding the control unit 130)
The control unit 130 is a controller. For example, various programs stored in a storage device inside the conversion device 100 are executed by a CPU (Central Processing Unit), MPU (Micro Processing Unit), etc., using the RAM as a work area. It is realized by being executed. Also, the control unit 130 is a controller, and is implemented by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array). The control unit 130 according to the embodiment includes, as shown in FIG. Realize or carry out.

(Regarding the identification unit 131)
The identifying unit 131 identifies related content that is related to the user to whom the provided content is provided. For example, in the example of FIG. 1, the identifying unit 131 collects viewing histories from the viewing content server 20 and identifies related content related to the user U1. Further, the specifying unit 131 collects posting histories from the input content server 30, and specifies related content that is related to the user U1. The specifying unit 131 then stores the specified related content in the related content storage unit 121 .

Further, the identifying unit 131 may identify the viewing content viewed by the user as the related content. For example, in the example of FIG. 1, the identification unit 131 identifies the viewing content viewed by the user U1 as the related content, and stores it in the related content storage unit 121 together with the corresponding viewing mode.

The identifying unit 131 may also identify input content input by the user as related content. For example, in the example of FIG. 1, the identification unit 131 identifies the input content input by the user U1 as the related content, and stores it in the related content storage unit 121 together with the corresponding input mode.

(Regarding the extraction unit 132)
The extraction unit 132 extracts, as easy content, content that is estimated to be easy to view from the viewing content, based on the manner in which the user views the viewing content. For example, in the example of FIG. 1, the extraction unit 132 extracts easy content based on user U1's operation on the viewing content.

Also, the extracting unit 132 may extract viewing content whose viewing time is less than a predetermined threshold as easy content. For example, in the example of FIG. 1, the extracting unit 132 selects viewing content whose time from being displayed on the screen of the terminal device 10 to performing a transition operation to another screen is equal to or less than a predetermined threshold as easy content. Extract.

The extraction unit 132 may also extract, as easy content, viewing content that satisfies a predetermined condition for details of operations performed during viewing. For example, in the example of FIG. 1, the extraction unit 132 extracts, as easy content, viewing content whose scroll speed during viewing by the user U1 is equal to or higher than a predetermined threshold.

Further, the extracting unit 132 may extract, as an easy part, a portion of the viewing content that is estimated to be easy to view, based on the manner in which the viewing content is viewed. For example, in the example of FIG. 1, the extraction unit 132 extracts the scroll speed is greater than or equal to a predetermined threshold is extracted as an easy portion.

Moreover, the extracting unit 132 may further extract content estimated to be not easy to view from the viewing content as non-easy content. For example, in the example of FIG. 1, the extraction unit 132 extracts difficult content based on user U1's operation on the viewing content.

Further, the extracting unit 132 may extract a portion of the viewing content that is estimated to be difficult to view as an uneasy portion based on the manner in which the viewing content is viewed. For example, in the example of FIG. 1, the extracting unit 132 extracts a range in which the scroll speed is equal to or less than a predetermined threshold value as an uneasy portion based on the scroll speed of the user U1 for each predetermined range of viewing content.

Further, the extracting unit 132 may extract easy-to-input content estimated to be easy to input from the input content based on the manner in which the user inputs the input content. For example, in the example of FIG. 1, the extracting unit 132 extracts input content whose input speed of the user U1 is equal to or higher than a predetermined threshold as easy-to-input content. Note that the extracting unit 132, for example, determines the input speed based on the time from when the content is distributed from the input content server 30 until the terminal device 10 transmits the input content to the input content server 30, the interruption time, the number of deletion operations, and the like. Easy-to-enter content may be extracted according to

Further, the extracting unit 132 may extract, as an easy-to-input portion, a portion of the input content that is estimated to be easy to input, based on the manner in which the input content is input. For example, in the example of FIG. 1, the extraction unit 132 calculates the input speed based on the input speed of the user U1 for each predetermined range of input content (for example, a range defined based on words, paragraphs, sentences, phrases, etc.). is greater than or equal to a predetermined threshold is extracted as an easy-to-input portion.

Moreover, the extraction unit 132 may further extract, from the input content, hard-to-input content estimated to be difficult to input. For example, in the example of FIG. 1, the extracting unit 132 extracts input content whose input speed of the user U1 is equal to or less than a predetermined threshold as content that is difficult to input.

Further, the extracting unit 132 may extract a portion of the input content that is estimated to be difficult to input as the difficult-to-input portion, based on the manner in which the input content was input. For example, in the example of FIG. 1, the extracting unit 132 extracts ranges in which the input speed is equal to or less than a predetermined threshold as input-difficult portions, based on user U1's input speed for each predetermined range of input content.

(Regarding the generation unit 133)
The generation unit 133 generates an expression model by learning the characteristics of the expressions of the related content identified by the identification unit 131 . For example, the generating unit 133 converts the related content identified by the identifying unit 131 into a vector using w2v, s2v, or the like, generates an expression model learned based on the vector, and stores it in the model storage unit 122. . For example, in the example of FIG. 1, the generation unit 133 converts words and sentences extracted from easy parts by techniques such as morphological analysis into vectors, and is an easy language model trained based on the vectors. When a word or sentence is input, an easy language model is generated that outputs a score indicating the degree of similarity with a group of words or sentences extracted from the easy part of the related content. In addition, the generation unit 133 converts the words and sentences extracted from the difficult part into vectors, and is a difficult language model trained based on the vector, and the provided content to be provided to the user U1 is input. Then, a difficult language model is generated that outputs a difficult part of the provided content that is estimated to be difficult for the user U1 to view.

In addition, the generation unit 133 generates an expression model so that, when predetermined content is input, content having a common expression with related content and having a similar intention to the input content is output. do. For example, when the provided content is input, the generation unit 133 outputs the provided content whose expression is converted so as to approach the expression of the easy content and the easy-to-input content, A representation model that has been learned such that the degree of similarity with content is equal to or greater than a predetermined threshold is generated.

Note that the generating unit 133 may analyze the related content using techniques such as morphological analysis and generate information about the thesaurus.

(Regarding the conversion unit 134)
The conversion unit 134 converts the expression of the provided content based on the expression of the related content identified by the identification unit 131 . For example, in the example of FIG. 1, the conversion unit 134 converts the presentation of the provided content based on the presentation of related content including the viewing content and the input content.

Further, the conversion unit 134 may convert the representation of the provided content so as to approximate the representation of the easy content extracted by the extraction unit 132 . For example, in the example of FIG. 1, the conversion unit 134 converts a vector corresponding to the viewing content that is estimated to be easily viewed by the user U1 among the viewing content specified as the related content, and a vector corresponding to the provided content after conversion. Transform the representation of the provided content so that it resembles a vector.

Further, the conversion unit 134 may convert the expression of the provided content so as to approximate the expression of the easy part extracted by the extraction unit 132 . For example, in the example of FIG. 1, the conversion unit 134 converts a vector corresponding to an easy part that is estimated to be easily viewed by the user U1 among the viewing contents specified as related contents, and a vector corresponding to the provided content after conversion. Transform the representation of the provided content so that it resembles a vector.

Further, the conversion unit 134 may further convert the representation of the provided content so as to move away from the representation of the difficult content extracted by the extraction unit 132 . For example, in the example of FIG. 1, the conversion unit 134 corresponds to a vector corresponding to viewing content that is estimated to be difficult for the user U1 to view, among the viewing content specified as related content, and the provided content after conversion. Transform the representation of the provided content so that it is not similar to the vector that

Further, the conversion unit 134 may convert the representation of the provided content so as to move away from the representation of the difficult part extracted by the extraction unit 132 . For example, in the example of FIG. 1, the conversion unit 134 converts a vector corresponding to a difficult portion that is estimated to be difficult for the user U1 to view among the viewing content specified as the related content, and the provided content after conversion. Transform the representation of the provided content so that the corresponding vectors are dissimilar.

Further, the conversion unit 134 may identify a portion of the provided content whose expression is similar to the difficult portion extracted by the extraction unit 132, and convert the expression of the identified portion. For example, in the example of FIG. 1, the conversion unit 134 identifies a portion of the provided content whose expression is similar to the difficult portion by comparing the vector corresponding to the provided content with the vector corresponding to the difficult portion. and convert the representation of that part.

Further, the conversion unit 134 may convert the expression of the provided content so as to approximate the expression of the easy-to-input content extracted by the extraction unit 132 . For example, in the example of FIG. 1, the conversion unit 134 converts a vector corresponding to the input content that is estimated to be input easily by the user U1 among the input content specified as the related content, and a vector corresponding to the provided content after conversion. Transform the representation of the provided content so that it resembles a vector.

Also, the conversion unit 134 may convert the expression of the provided content so as to approximate the expression of the easy-to-input portion extracted by the extraction unit 132 . For example, in the example of FIG. 1, the conversion unit 134, of the input content specified as related content, corresponds to a vector corresponding to an easy-to-input portion that is estimated to be easy for user U1 to input, and a provided content after conversion. Transform the representation of the provided content so that it is similar to the vector that

Further, the conversion unit 134 may convert the expression of the provided content so as to move away from the expression of the difficult-to-input content extracted by the extraction unit 132 . For example, in the example of FIG. 1, the conversion unit 134 corresponds to a vector corresponding to the input content that is estimated to be difficult for the user U1 to input among the input contents specified as the related content, and the provided content after conversion. Transform the representation of the provided content so that it is not similar to the vector that

Further, the conversion unit 134 may convert the expression of the provided content so as to move away from the expression of the difficult-to-input portion extracted by the extraction unit 132 . For example, in the example of FIG. 1, the conversion unit 134 converts a vector corresponding to an input-difficult part that is estimated to be difficult for the user U1 to input from among the input content specified as related content, and the provided content after conversion. Transform the representation of the provided content so that it is dissimilar to the vectors corresponding to .

Further, the conversion unit 134 may identify a portion of the provided content whose expression is similar to the difficult-to-input portion extracted by the extraction unit 132, and convert the expression of the identified portion. For example, in the example of FIG. 1, the conversion unit 134 compares the vector corresponding to the provided content with the vector corresponding to the hard-to-input portion, and converts the portion of the provided content that is similar in expression to the difficult-to-input portion. , and transform the representation of that part.

Further, the conversion unit 134 may convert the expression of the provided content so as not to change the intention of the provided content before conversion of the expression. For example, the conversion unit 134 converts the provided content such that the vector corresponding to the provided content before conversion is similar to the vector corresponding to the provided content after conversion.

Also, the conversion unit 134 may convert the representation of the provided content using the representation model. For example, in the example of FIG. 1, the conversion unit 134 inputs words and sentences extracted from the provided content by techniques such as morphological analysis into the difficult language model, and presumes that viewing of the provided content is difficult for the user U1. Extract the non-easy parts that are Then, the conversion unit 134 uses the thesaurus to extract synonyms and similar words of words and sentences extracted from the difficult part of the provided content as candidates, and among the candidates, the score output by the easy language model is the highest. High candidates are used to transform the representation of the difficult part of the provided content.

[4. Flow of conversion process]
Here, the procedure of conversion processing of the conversion device 100 according to the embodiment will be described with reference to FIG. FIG. 6 is a flowchart illustrating an example of a procedure of conversion processing according to the embodiment.

As shown in FIG. 6, the conversion device 100 identifies related content (step S101). For example, in the example of FIG. 1, the conversion device 100 collects viewing histories from the viewing content server 20 and identifies related content related to the user U1. Conversion device 100 also collects posting histories from input content server 30 and identifies related content that is related to user U1.

Subsequently, the conversion device 100 extracts easy content and non-easy content from the related content (step S102). For example, in the example of FIG. 1, the conversion device 100 extracts the easy part and the difficult part of the related content, and creates an easy language model that has learned the easy part and a difficult language model that has learned the difficult part. Generate.

Subsequently, the conversion device 100 determines whether or not a request for providing content to be provided has been received from the terminal device 10 (step S103). If no provision request has been received (step S103; No), the conversion device 100 repeats the process of identifying related content and the process of extracting easy content and non-easy content.

On the other hand, if a provision request has been accepted (step S103; Yes), the conversion device 100 provides content to be provided whose expression has been converted so as to approximate the expression of easy content and away from the expression of difficult content (step S104), End the process.

[5. Modification]
The above-described embodiment is an example, and various modifications and applications are possible.

[5-1. Regarding the relationship between "easy" and "not easy"]
In the above-described embodiment, the extracting unit 132 treats viewing content with a fast scroll speed and input content with a fast input speed as easy content, and determines viewing content with a slow scroll speed and input content with a slow input speed as non-easy content. However, the easy content and the non-easy content may have any relationship. For example, the extracting unit 132 determines viewing content whose scrolling speed is equal to or higher than a predetermined threshold value and input content whose input speed is equal to or higher than a predetermined threshold among the related content as easy content, and regards all other related content as non-easy content. It can also be used as content. Further, for example, the extracting unit 132 may regard viewing content with a scroll speed equal to or higher than a first threshold as easy content, and may regard viewing content with a scroll speed equal to or lower than a second threshold as non-easy content. Then, the extracting unit 132 may regard input content whose input speed is equal to or higher than the first threshold among the related content as easy content, and may regard input content whose input speed is equal to or lower than the second threshold as non-easy content.

[5-2. Conversion of provided content]
In the above-described embodiment, an example was given in which the conversion unit 134 converts words or sentences extracted from the difficult part of the provided content into synonyms or similar words, but the function of the conversion unit 134 is limited to such a configuration. not. For example, the conversion unit 134 may convert one word into multiple words, convert multiple words into one word, or change kanji into hiragana. Further, for example, the conversion unit 134 may change the word order of the sentence extracted as the difficult part or the length of the sentence. In addition, if the provided content includes expressions unique to the region (for example, dialects) different from the standard language, the conversion unit 134 may convert the provided content according to the region where the user is based.

Further, in the above-described embodiment, an example was given in which the conversion unit 134 converts provided content that is assumed to be difficult for the user to view, but the function of the conversion unit 134 is not limited to such a configuration. For example, when it is estimated that the provided content is easy for the user to view, the conversion unit 134 approximates the expression to an expression according to the user's high vocabulary and the user's degree of understanding of the category to which the provided content belongs. The representation of the provided content may be converted as follows.

Further, in the above-described embodiment, an example in which the conversion unit 134 converts words and sentences included in the provided content has been described, but the function of the conversion unit 134 is not limited to such a configuration. For example, the conversion unit 134 may convert the structure of sentences and images in the provided content that is estimated to be difficult for the user to view so as to approximate the structure of the easy content or the easy part. In addition, when the provided content that is estimated to be difficult for the user to view is audio-related content, the conversion unit 134 converts the audio so as to approach the expression of the easy content or the easy part that is estimated to be easy for the user to view. to convert

Note that the conversion device 100 may provide the terminal device 10 with provision content in which the portion converted by the conversion unit 134 is highlighted. The conversion device 100 may also provide the terminal device 10 with provision content in which the words and sentences converted by the conversion unit 134 are listed and displayed.

[6. effect〕
As described above, the conversion device 100 according to the embodiment has the identification unit 131, the extraction unit 132, the generation unit 133, and the conversion unit . The identifying unit 131 identifies related content that is related to the user to whom the provided content is provided. Further, the identifying unit 131 identifies viewing content viewed by the user as related content. The extraction unit 132 extracts, as easy content, content that is estimated to be easy to view from the viewing content, based on the manner in which the user views the viewing content. The generation unit 133 generates an expression model by learning the characteristics of the expressions of the related content identified by the identification unit 131 . The conversion unit 134 converts the expression of the provided content based on the expression of the related content identified by the identification unit 131 .

As a result, the conversion device 100 according to the embodiment converts the expression of the content to be provided based on the expression of the content related to the user, so that it is possible to provide the user with the content with an appropriate expression.

Further, in the conversion device 100 according to the embodiment, for example, the extraction unit 132 extracts viewing content whose viewing time is less than a predetermined threshold as easy content. Further, the extracting unit 132 extracts, as easy content, viewing content that satisfies a predetermined condition in terms of operation details during viewing. Then, the conversion unit 134 converts the expression of the provided content so as to approximate the expression of the easy content extracted by the extraction unit 132 . Moreover, the extracting unit 132 extracts a portion of the viewing content that is estimated to be easy to view as an easy portion, based on the manner in which the viewing content is viewed. Then, the conversion unit 134 converts the expression of the provided content so as to approximate the expression of the easy part extracted by the extraction unit 132 .

As a result, the conversion device 100 according to the embodiment can bring the expression of the provided content closer to the expression of the easy content or the easy part, so that it can provide the user with content of appropriate expression.

In addition, in the conversion device 100 according to the embodiment, for example, the extracting unit 132 further extracts content estimated to be difficult to view from the viewing content as non-easy content. Then, the conversion unit 134 further converts the representation of the provided content so as to move away from the representation of the difficult content extracted by the extraction unit 132 . Moreover, the extracting unit 132 extracts a portion of the viewing content that is estimated to be difficult to view as an uneasy portion based on the manner in which the viewing content is viewed. Then, the conversion unit 134 converts the expression of the provided content so as to move away from the expression of the difficult part extracted by the extraction unit 132 . Further, the conversion unit 134 identifies a portion of the provided content whose expression is similar to the difficult portion extracted by the extraction unit 132, and converts the expression of the identified portion.

As a result, the conversion device 100 according to the embodiment can keep the expression of the provided content away from the expression of the difficult content or the difficult part, so that it is possible to provide the user with the content of appropriate expression.

Further, in the conversion device 100 according to the embodiment, for example, the identifying unit 131 identifies input content input by the user as related content. Further, the extracting unit 132 extracts easy-to-input content estimated to be easy to input from among the input content, based on the manner in which the user inputs the input content. Then, the conversion unit 134 converts the expression of the provided content so as to approximate the expression of the easy-to-input content extracted by the extraction unit 132 . Further, the extracting unit 132 extracts a portion of the input content that is estimated to be easy to input as an easy-to-input portion, based on the manner in which the input content is input. Then, the conversion unit 134 converts the expression of the provided content so as to approximate the expression of the easy-to-input portion extracted by the extraction unit 132 .

As a result, the conversion device 100 according to the embodiment can bring the expression of the provided content closer to the expression of the easy-to-input content or the easy-to-input portion, so that it is possible to provide the user with content of appropriate expression.

In addition, in the conversion device 100 according to the embodiment, for example, the extracting unit 132 further extracts, from among the input contents, hard-to-input contents estimated to be not easy to input. Then, the conversion unit 134 converts the expression of the provided content so as to move away from the expression of the hard-to-input content extracted by the extraction unit 132 . Moreover, the extracting unit 132 extracts a portion of the input content that is estimated to be difficult to input as an input-hard-to-input portion, based on the manner in which the input content was input. Then, the conversion unit 134 converts the expression of the provided content so as to move away from the expression of the difficult-to-input portion extracted by the extraction unit 132 . In addition, the conversion unit 134 identifies a portion of the provided content whose expression is similar to the difficult-to-input portion extracted by the extraction unit 132, and converts the expression of the identified portion.

As a result, the conversion device 100 according to the embodiment can keep the representation of the provided content away from the representation of the content that is difficult to input or the part that is difficult to input, so that it is possible to provide the user with content that is appropriately represented. .

Further, in the conversion device 100 according to the embodiment, for example, the conversion unit 134 converts the expression of the provided content so as not to change the intention of the provided content before expression conversion.

As a result, the conversion device 100 according to the embodiment converts the expression without changing the intention of the content, so it is possible to provide the user with content with an appropriate expression.

Also, in the conversion device 100 according to the embodiment, for example, the conversion unit 134 converts the representation of the provided content using the representation model. In addition, the generation unit 133 generates an expression model so that, when predetermined content is input, content having a common expression with related content and having a similar intention to the input content is output. do.

As a result, the conversion device 100 according to the embodiment performs conversion processing using a model that has learned content related to the user, and therefore can provide content with appropriate expressions to the user.

[7. Hardware configuration]
Also, the conversion apparatus according to each of the embodiments described above is implemented by a computer 1000 configured as shown in FIG. 7, for example. The conversion device 100 will be described below as an example. FIG. 7 is a hardware configuration diagram showing an example of a computer that implements the functions of the conversion device. Computer 1000 has CPU 1100 , ROM 1200 , RAM 1300 , HDD 1400 , communication interface (I/F) 1500 , input/output interface (I/F) 1600 and media interface (I/F) 1700 .

The CPU 1100 operates based on programs stored in the ROM 1200 or HDD 1400 and controls each section. The ROM 1200 stores a boot program executed by the CPU 1100 when the computer 1000 is started, programs dependent on the hardware of the computer 1000, and the like.

The HDD 1400 stores programs executed by the CPU 1100, data used by the programs, and the like. Communication interface 1500 receives data from another device via communication network 500 (corresponding to network N in the embodiment) and sends it to CPU 1100, and also transmits data generated by CPU 1100 via communication network 500 to other devices. Send to device.

The CPU 1100 controls output devices such as displays and printers, and input devices such as keyboards and mice, through an input/output interface 1600 . CPU 1100 acquires data from an input device via input/output interface 1600 . Also, CPU 1100 outputs the generated data to an output device via input/output interface 1600 .

Media interface 1700 reads programs or data stored in recording medium 1800 and provides them to CPU 1100 via RAM 1300 . CPU 1100 loads such a program from recording medium 1800 onto RAM 1300 via media interface 1700, and executes the loaded program. The recording medium 1800 is, for example, an optical recording medium such as a DVD (Digital Versatile Disc) or a PD (Phase change rewritable disc), a magneto-optical recording medium such as an MO (Magneto-Optical disk), a tape medium, a magnetic recording medium, or a semiconductor memory. etc.

For example, when the computer 1000 functions as the conversion device 100 , the CPU 1100 of the computer 1000 implements the functions of the control unit 130 by executing a program loaded onto the RAM 1300 . Further, each data in the storage device of the conversion device 100 is stored in the HDD 1400 . CPU 1100 of computer 1000 reads these programs from recording medium 1800 and executes them, but as another example, these programs may be obtained from another device via a predetermined communication network.

[8. others〕
As described above, some of the embodiments of the present application have been described in detail based on the drawings. It is possible to carry out the invention in other forms with modifications.

Further, the configuration of the conversion device 100 described above can be flexibly changed, for example, by calling an external platform or the like using an API (Application Programming Interface), network computing, or the like, depending on the function.

In addition, the "unit" described in the claims can be read as "means", "circuit", or the like. For example, the identifying unit can be read as identifying means or a specific circuit.

10 terminal device 20 viewing content server 30 input content server 100 conversion device 110 communication unit 120 storage unit 121 related content storage unit 122 model storage unit 130 control unit 131 identification unit 132 extraction unit 133 generation unit 134 conversion unit

Claims (11)

a specifying unit that specifies viewing content viewed by the user as related content related to the user to whom the provided content is provided;
Based on the manner in which the user views the viewing content, content that is estimated to be easy to view is extracted as easy content from the viewing content, and viewing is difficult from the viewing content. a first extraction unit that extracts content estimated to be as non-easy content;
generating an easy language model trained to output a degree of similarity with easy content when content is input based on the easy content; a generation unit that generates a difficult language model that has been trained to output a difficult part of the content that is estimated to be difficult for the user to view when the content is viewed;
Using the difficult language model, a difficult part is extracted from the provided content, and among a plurality of candidates obtained by converting the expression of the difficult part, a candidate with the highest similarity output by the easy language model is selected. and a conversion unit that converts the expression of the difficult part using the conversion device. The first extraction unit is
2. The conversion device according to claim 1 , wherein viewing content whose viewing time is less than a predetermined threshold is extracted as said easy content. The first extraction unit is
3. The conversion device according to claim 1 , wherein viewing contents satisfying a predetermined condition for details of operation during viewing are extracted as said easy contents. The identification unit
4. The conversion device according to any one of claims 1 to 3 , wherein input content input by said user is specified as said related content. a second extraction unit that extracts easy-to-input content that is estimated to be easy to input from among the input content based on the manner in which the input content is input by the user;
The conversion unit
5. The conversion device according to claim 4 , wherein the expression of the provided content is converted so as to approximate the expression of the easy-to-input content extracted by the second extraction unit. The second extractor is
extracting a portion of the input content that is presumed to be easy to input as an easy-to-input portion, based on the manner in which the input content is input;
The conversion unit
6. The conversion device according to claim 5 , wherein the expression of the provided content is converted so as to approximate the expression of the easy-to-input portion extracted by the second extraction unit. The second extractor is
Furthermore, the conversion unit extracts content that is estimated to be difficult to input from the input content,
7. The conversion device according to claim 5 , wherein the expression of the provided content is converted so as to move away from the expression of the hard-to-input content extracted by the second extraction unit. The second extractor is
extracting a portion of the input content that is presumed to be difficult to input as a non-easy-to-input portion, based on the manner in which the input content is input;
The conversion unit
8. The conversion device according to claim 7 , wherein the expression of the provided content is converted so as to move away from the expression of the difficult-to-input portion extracted by the second extraction unit. The conversion unit
9. The conversion device according to claim 8 , wherein a part whose expression is similar to the hard-to-input part extracted by the second extraction unit is specified in the provided content, and the expression of the specified part is converted. . A conversion method executed by a conversion device,
an identifying step of identifying viewing content viewed by the user as related content related to the user to whom the provided content is provided;
Based on the manner in which the user views the viewing content, content that is estimated to be easy to view is extracted as easy content from the viewing content, and viewing is difficult from the viewing content. a first extracting step of extracting content estimated to be as non-easy content;
generating an easy language model trained to output a degree of similarity with easy content when content is input based on the easy content; a generation step of generating a difficult language model that has been trained to output a difficult part of the content that is estimated to be difficult for the user to view when the content is viewed;
Using the difficult language model, a difficult part is extracted from the provided content, and among a plurality of candidates obtained by converting the expression of the difficult part, a candidate with the highest similarity output by the easy language model is selected. a conversion step of converting the representation of the difficult part using : an identification procedure for identifying the viewing content viewed by the user as related content related to the user to whom the provided content is provided;
Based on the manner in which the user views the viewing content, content that is estimated to be easy to view is extracted as easy content from the viewing content, and viewing is difficult from the viewing content. a first extraction procedure for extracting content estimated to be as non-easy content;
generating an easy language model trained to output a degree of similarity with easy content when content is input based on the easy content; a generation procedure for generating a difficult language model that has been trained to output a difficult part of the content that is estimated to be difficult for the user to view, when
Using the difficult language model, a difficult part is extracted from the provided content, and among a plurality of candidates obtained by converting the expression of the difficult part, a candidate with the highest similarity output by the easy language model is selected. A conversion program for causing a computer to execute a conversion procedure for converting the representation of the difficult part using a conversion procedure.

Images