JP2018005324A - Generation apparatus, generation method, and generation program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To generate an appropriate summary according to a user.SOLUTION: A generation apparatus 100 includes an acquisition unit 131 and a generation unit 134. The acquisition unit 131 acquires character information and user-related information which is information on a recipient user. The generation unit 134 generates summary information from the character information acquired by the acquisition unit 131, on the basis of the user-related information acquired by the acquisition unit 131. For example, the generation unit 134 generates summary information from the character information, on the basis of a model learned by information formed by associating first character information with second character information which is summary information of the first character information.SELECTED DRAWING: Figure 2

Description

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

  2. Description of the Related Art Conventionally, a technique relating to summary generation that automatically generates a summary of a sentence has been provided. For example, a technique for generating a title from an article or a headline is provided.

JP 2014-056499 A

  However, in the above-described conventional technology, it is not always possible to generate an appropriate summary according to the user. For example, when the same summary is distributed to a plurality of users whose terminal devices used by each user have different sizes, fonts, etc., the summary is not necessarily the best for each user's terminal device, and is appropriate for the user. Is not necessarily generated.

  This application is made in view of the above, Comprising: It aims at providing the production | generation apparatus, the production | generation method, and production | generation program which can produce | generate an appropriate summary according to a user.

  The generation device according to the present application is acquired by the acquisition unit based on the acquisition unit that acquires character information and user-related information that is information related to a user who is a providing destination, and the user-related information acquired by the acquisition unit. And a generation unit that generates summary information from character information.

  According to one aspect of the embodiment, there is an effect that an appropriate summary can be generated according to the user.

FIG. 1 is a diagram illustrating an example of a generation process according to the embodiment. FIG. 2 is a diagram illustrating a configuration example of the generation apparatus according to the embodiment. FIG. 3 is a diagram illustrating an example of the editing information storage unit according to the embodiment. FIG. 4 is a diagram illustrating an example of a model information storage unit according to the embodiment. FIG. 5 is a diagram illustrating an example of a user-related information storage unit according to the embodiment. FIG. 6 is a diagram illustrating an example of a content information storage unit according to the embodiment. FIG. 7 is a flowchart illustrating an example of model learning according to the embodiment. FIG. 8 is a flowchart illustrating an example of generation of a shortened title according to the embodiment. FIG. 9 is a diagram illustrating an example of generation of a shortened title according to the embodiment. FIG. 10 is a diagram illustrating an example of model generation using user-related information according to the embodiment. FIG. 11 is a diagram illustrating an example of generation of a two-stage summary according to the embodiment. FIG. 12 is a diagram illustrating an example of generation of a shortened title according to the terminal device according to the embodiment. FIG. 13 is a hardware configuration diagram illustrating an example of a computer that realizes the function of the generation apparatus.

  Hereinafter, a generation apparatus, a generation method, and a mode for executing a generation program (hereinafter referred to as “embodiment”) according to the present application will be described in detail with reference to the drawings. Note that the generation device, the generation method, and the generation program according to the present application are not limited by this embodiment. In the following embodiments, the same portions are denoted by the same reference numerals, and redundant description is omitted.

(Embodiment)
[1. Generation process)
First, an example of the generation process according to the embodiment will be described with reference to FIG. FIG. 1 is a diagram illustrating an example of a generation process according to the embodiment. Note that the generation processing in the present embodiment includes processing for distributing content including the generated shortened title. The generation device 100 illustrated in FIG. 1 provides a distribution service that distributes content including a shortened title based on a distribution request from the terminal device 10.

  As illustrated in FIG. 1, the generation system 1 includes a terminal device 10 and a generation device 100. The terminal device 10 and the generation device 100 are connected to be communicable by wire or wireless via a predetermined communication network (not shown). The generation system 1 illustrated in FIG. 1 may include a plurality of terminal devices 10 and a plurality of generation devices 100.

  The terminal device 10 is an information processing device used by a user. The terminal device 10 is realized by, for example, a smartphone, a tablet terminal, a notebook PC (Personal Computer), a desktop PC, a mobile phone, a PDA (Personal Digital Assistant), or the like. In the example illustrated in FIG. 1, the terminal device 10 is a smartphone having a touch panel function. Hereinafter, the terminal device 10 may be referred to as a user. That is, hereinafter, the user can be read as the terminal device 10.

  Further, the terminal device 10 requests the generation device 100 to distribute content. In addition, the terminal device 10 displays the content distributed from the generation device 100 using a predetermined application. The predetermined application may be a browser.

  The generation device 100 is an information processing device that provides a service for distributing content including a shortened title to the terminal device 10. Moreover, the production | generation apparatus 100 produces | generates summary information from character information based on user related information. For example, the generating apparatus 100 generates a shortened title obtained by further shortening the title of the predetermined content based on the user related information.

  The generation system 1 may include a provider device (not shown) and an editor terminal (not shown). For example, the provider device is an information processing device used by a content provider, a so-called content provider or the like. For example, the providing device submits content created by the providing device to the generating device 100. Further, for example, the editor terminal is an information processing apparatus used by a predetermined editor. For example, the editor terminal is a terminal device used by an editor who creates an abbreviated title from the text or title of content, for example.

  Hereinafter, generation and distribution of a shortened title by the generation apparatus 100 will be described with reference to FIG. In the example of FIG. 1, it is assumed that the generation apparatus 100 has already acquired the summary list EL11 in which a shortened title and an original title are associated with each other from an editor terminal used by a predetermined editor. The summary list EL11 shown in FIG. 1 corresponds to the editing information stored in the editing information storage unit 121 shown in FIG. 3, and for each content, a shortened title created by a predetermined editor, the original title of the content, Is associated information. Specifically, in the summary list EL11, a shortened title A is created from a title A of predetermined content, a shortened title B is created from a title B of predetermined content, and a shortened title C is created from a title C of predetermined content. Indicates that In the example of FIG. 1, in order to simplify the description, the summary list EL <b> 11 may include the article itself of the content associated with the shortened title.

  First, the generation device 100 generates a summary generation model M11 (hereinafter, also simply referred to as “model M11”) from the summary list EL11 by learning using the summary list EL11 (step S11). For example, the generation apparatus 100 acquires information shown in the summary list EL11 from the editing information storage unit 121 (see FIG. 3), and learns the summary generation model M11. As described above, in the example of FIG. 1, the generation apparatus 100 learns the summary generation model M11 using a combination of the abbreviated title created by a predetermined editor included in the summary list EL11 and the original title. For example, the generation apparatus 100 uses a combination of a shortened title created by a predetermined editor included in the summary list EL11 and the title as the original as a positive example, and character information such as a predetermined title is input. To learn the summary generation model M11.

  Further, the generation device 100 stores the summary generation model M11 generated by learning in the model information storage unit 122 (see FIG. 4). For example, the generation apparatus 100 learns the summary generation model M11 using various conventional techniques such as a neural network such as a current neural network and dependency analysis based on the original title and the abbreviated title included in the summary list EL11. To do. The summary generation model M11 is a model for generating a shortened title within a predetermined number of characters in response to, for example, content titles and article inputs. Details of the summary generation model M11 will be described later.

  Then, the generation device 100 acquires a content distribution request from the terminal device 10 (step S12). In the example of FIG. 1, the generation device 100 acquires user-related information regarding the user U1 who uses the terminal device 10 from the terminal device 10 in step S12. For example, the generation apparatus 100 acquires various pieces of information such as attribute information of the user U1, terminal information about the terminal apparatus 10, and context information of the user U1 from the terminal apparatus 10. In the example of FIG. 1, the generation apparatus 100 acquires user-related information as shown in the user-related information storage unit 123.

  For example, the generation apparatus 100 acquires attribute information indicating that the user U1 is a man in his 30s and the residence is “Area A” as the user-related information. Further, for example, as the user related information, the generation apparatus 100 has the model number of the terminal device 10 used by the user U1 as “TM11”, the screen size as “SZ11”, and the set font as “FT11”. The terminal information indicating that is acquired. The font “FT11” includes various types of information such as the font type and font size. Further, for example, the generation apparatus 100 indicates that the current position of the terminal apparatus 10 used by the user U1 is “LC11”, the acceleration is “AC11”, and the context is “CN11” as user-related information. Get context information. The user related information may be any information as long as it is information related to the user. For example, the user-related information may be a user's brain wave, heart rate, estimated emotion information, or the like. In addition, the context may be divided into a plurality of types depending on the granularity, such as during a long vacation, traveling, or in a train. The user-related information here may include various information such as the environment surrounding the user, and may include various information such as weather, temperature, humidity, event, stock price, and news of the day. .

  The context here means the situation of the user U1 and the terminal device 10 and the environment (background) of the user U1 and the terminal device 10. For example, the context is information corresponding to a situation where the terminal device 10 is used by the user U1 and a state where the user U1 who owns the terminal device 10 is placed. That is, the context in FIG. 1 corresponds to the situation of the user U1, for example, the environment where the user U1 is. Specifically, the context in FIG. 1 corresponds to the situation of the user U1 at the time when each user related information is acquired.

  Further, the fact that the context of the user U1 is “CN11” means that information indicating that the current position of the terminal device 10 used by the user U1 is “LC11”, information indicating that the acceleration is “AC11”, It may be estimated based on various information.

  Further, in the example of FIG. 1, in order to simplify the description, a case where the generation device 100 acquires various types of user-related information in step S <b> 12 is illustrated. You may acquire at the timing of. For example, the generation device 100 may acquire the context information of the user U1 at a predetermined timing and estimate the context of the user U1 based on the acquired context information. Further, when the attribute information of the user U1 has been acquired before step S12, the generation apparatus 100 does not have to acquire the attribute information of the user U1 in step S12. In this case, the generation device 100 may acquire information for identifying that the user who uses the terminal device 10 is the user U1.

  Below, it demonstrates as what the production | generation apparatus 100 delivers content CT11-CT14 with respect to the delivery request | requirement in step S12. Here, the content CT11 means a content identified by the content ID “CT11”. The content CT12 means a content identified by the content ID “CT12”. The content list CL11 illustrated in FIG. 1 corresponds to the content information stored in the content information storage unit 124 illustrated in FIG. For example, the content CT11 includes the title “X month Y day XX prefecture bus rollover, immediately preceding speed, over 95 km / h”, image IM11, and the like. The content CT11 includes articles, but is not shown in FIG.

  The generating apparatus 100 generates the shortened titles AT11 to AT14 corresponding to each of the contents CT11 to CT14 in order to distribute the contents CT11 to CT14 (step S13). For example, the generating apparatus 100 generates the abbreviated titles AT11 to AT14 as shown in the title list AL11 based on the titles corresponding to the contents CT11 to CT14.

  In FIG. 1, in order for the generation apparatus 100 to distribute the contents CT11 to CT14, the abbreviated titles AT11 to AT14 corresponding to the contents CT11 to CT14 are generated. In the following, the abbreviated title AT11 corresponding to the contents CT11 is generated. Will be described as an example.

  Here, the generation apparatus 100 uses the summary generation model M11 in order to generate a shortened title. For example, the generation device 100 calculates a score at each position of each word using the summary generation model M11. For example, the generation device 100 calculates the score of each word at each position according to the following equation (1).

  W_x score = P (W_x | W_ {1} ^ {x−1}, INPUT) (1)

  In the above formula (1), “W_x” indicates an arbitrary word. In the above formula (1), “W_ {1} ^ {x−1}” indicates the first to x−1th words before the word “W_x”. Thus, in the above formula (1), all words output before the character to be estimated (word “W_x”) are used as conditions. Also, “x” in “W_x” and “x−1” in “W_ {1} ^ {x−1}” indicate the positional relationship of the corresponding words, and “W_ {x−1}” is The word is located immediately before “W_x”. That is, “W_x” means a word following “W_ {x−1}”.

  “INPUT” in the right side of the above equation (1) indicates a character string to be input to the model M11, for example. For example, in the example of FIG. 1, the title of the content CT11 is input. Note that the word corresponding to “W_x” and the word corresponding to “W_ {x−1}” may be different words or the same word. Further, the word is not limited to the word included in the input “INPUT” but may be various words. For example, the word “W” may be an abbreviation, abbreviation or nickname of the word included in the input. In addition, the word here may include a morpheme, a character (Hiragana, Katakana, one Kanji character) and the like.

  For example, the score of W_x is the probability of transitioning to “W_x” next to an arbitrary word “W_ {x−1}” when generating a summary (short title) using an arbitrary character string “INPUT” as an input. The corresponding score. For example, the score of W_x may be a score based on the posterior probability. Thus, “P” in the right side of the above formula (1) is a function that returns a predetermined score according to an arbitrary character string “INPUT”.

  For example, the score of an arbitrary word “W1” positioned at the first character is calculated by the following equation (2).

  W1 score = P (W1 | <S>, INPUT) (2)

  In the above formula (2), “<S>” in the right side is a symbol indicating the head. That is, “P” in the right side of the above formula (2) is the probability that an arbitrary word “W1” will appear at the beginning when an abstract (short title) is generated by inputting an arbitrary character string “INPUT”. This function returns the corresponding score.

  For example, the following formula (3) indicates a score corresponding to the probability that the word “bus” appears at the head when generating a summary (short title) using an arbitrary character string “INPUT” as an input. In this case, it is assumed that the arbitrary character string “INPUT” is the title of the content CT11 “X month Y day XX prefecture bus rollover, immediately preceding speed, over 95 km / h”.

  W1 (“Bus”) score = P (Bus | <S>, INPUT) (3)

  Also, for example, the following equation (4) corresponds to the probability of transition to the word “overturn” next to the word “bus” when generating a summary (short title) with an arbitrary character string “INPUT” as input. The score to be shown. In this case, it is assumed that the arbitrary character string “INPUT” is the title of the content CT11 “X month Y day XX prefecture bus rollover, immediately preceding speed, over 95 km / h”.

  W2 ("rollover") score = P (rollover | bus, INPUT) (4)

  In order to simplify the explanation, the above formula (4) shows a case where the W1 score is not used for calculating the W2 score. However, the W2 score obtained by adding or multiplying the W1 score is used. It may be calculated. That is, the score of W2 may be a score that takes into account the influence of the score of W1.

  As described above, for example, a tree structure expanded with all words in an arbitrary character string “INPUT” is based on a tree structure having each word as a node (see “Tree Structure ST11” in FIG. 9; the same applies hereinafter). Indicated. As described above, the generation apparatus 100 calculates a score based on the conditional probability of each word and connects the words based on the score, as shown in the examples of the above formulas (3) and (4). Generate a short title. For example, a shortened title may be generated by connecting words of a route having the maximum total score of nodes (words) corresponding to a predetermined sequence (route) of the tree structure ST11 as shown in FIG.

  In the example of FIG. 1, it is assumed that the upper limit value of the number of characters of the shortened title is determined by the screen size “SZ11” and the font “FT11” of the terminal device 10. For example, the upper limit of the number of characters of the shortened title is “13 full-width characters”. For example, when the font is “FT11”, the number of characters that can be displayed in the display area AR11 of the terminal device 10 is 13 full-width characters.

  For example, the generation apparatus 100 sets a route having the maximum score among the routes in the tree structure that satisfies the upper limit “13 full-width characters” as the abbreviated title. For example, when the number of characters exceeds the upper limit value by adding a node in a character string in which words corresponding to a predetermined path of the tree structure are added, the generation apparatus 100 does not search for that node, The search for a further route may be terminated. The generation apparatus 100 appropriately uses various conventional techniques such as beam search to suppress an increase in the search range, which will be described in detail later.

  Here, in the example illustrated in FIG. 1, the generation device 100 calculates a score in consideration of user related information. For example, the generation device 100 calculates a score that takes into account the user-related information of the user U1 acquired in step S12. For example, the generating apparatus 100 calculates a score that takes user-related information into account using the following equation (5).

  Score of W_x = P (W_x | W_ {1} ^ {x-1}, INPUT) * U (Wx | W_ {1} ^ {x-1}, USER) (5)

  In the right side of the above formula (5), “U” is a word “W_ {x−1}” next to an arbitrary word “W_ {x−1}” when a short title is generated for the user related information “USER” of an arbitrary user. The function returns a score corresponding to the probability of transition to “W_x”. For example, “U” in the right side of the equation (5) may be a function that returns a score based on the posterior probability. Thus, “U” in the right side of the above formula (5) is a function that returns a predetermined score according to user-related information “USER” of an arbitrary user. The user related information “USER” includes user attribute information, terminal information, and context information. The above equation (5) is an example of calculating the score of W_x, and the generating apparatus 100 may calculate the score of W_x by various methods such as addition. For example, the generating apparatus 100 may calculate the score of W_x using the following equation (6). For example, the function “f” in the following equation (6) is obtained by adding the value of the function “P” or the value of the function “U”, such as adding the value of the function “P” or the value of the function “U”. It is a function used to calculate a predetermined score.

  Score of W_x = f (P (W_x | W_ {1} ^ {x-1} Wx-1, INPUT), U (Wx | W_ {1} ^ {x-1} Wx-1, USER)) (6)

  For example, the function “U” may be realized by a predetermined user model. For example, when a predetermined shortened title is distributed to each user, when the user selects the shortened title, the user model is obtained by learning the shortened title and the user related information of the user as positive examples. It may be generated. In this case, for example, each feature of the user model includes features related to user-related information such as user attribute information, terminal information, and context information. Note that the above is an example, and the function “U” may be generated in any manner. For example, when the screen size of the terminal information is small, the function “U” may calculate the score so that the score of a word with a small number of characters is high. Further, for example, the generation apparatus 100 may increase the score of words in news in the user's news browsing history. For example, the generation apparatus 100 may use a language model of news that the user is browsing as in the above formula (5), or simply set the function “U” to “U (W_x | USER)” or the like. You may make it raise the probability of the word which it knows (feels naturally). Further, the generation apparatus 100 sets the function “U” as “U () = U 1 () * U 2 () *...”, Etc., the function “U 1” is a score of a user group of a certain age, and the function “U 2” is a certain region. The score of the function “U” may be calculated using a score calculated for each content of the user-related information, such as the score of the user layer.

  For example, the following formula (7) indicates a score corresponding to the probability that the word “bus” appears at the head in the case of generating a shortened title for the user related information “U1” when the user is U1. . In this case, it is assumed that the arbitrary character string “INPUT” is the title of the content CT11 “X month Y day XX prefecture bus rollover, immediately preceding speed, over 95 km / h”.

  Score of W1 (“bus”) = P (bus | <S>, INPUT) * U (bus | <S>, U1) (7)

  The above equation (7) indicates a score corresponding to the probability that the word “bus” appears at the head when the user is U1 and an abstract (short title) is generated by inputting an arbitrary character string “INPUT”. . For example, in the example shown in FIG. 9, the score corresponding to the probability that the word “bus” appears at the head is “0.6”. For example, in the example illustrated in FIG. 9, the score corresponding to the probability that the word “◯◯” appears at the top is “0.62”.

  As described above, in the example of FIG. 1, the upper limit value of the number of characters of the abbreviated title is “full-width 13” characters, and the generation apparatus 100 has a tree structure ST11 that satisfies the upper limit value of the number of characters “full-width 13” (see FIG. 9). Among the routes in the middle, the route with the highest score is set as a shortened title. For example, the generation device 100 searches the tree structure ST11 (see FIG. 9) corresponding to the title of the content CT11, thereby shortening the character string connecting the words corresponding to the route having the maximum score within 13 full-width characters. Generate as a title.

  For example, as illustrated in the title list AL11 of FIG. 1, the generation device 100 generates a shortened title AT11 corresponding to the title of the content CT11. Specifically, the generation device 100 generates a shortened title AT11 corresponding to the title of the content CT11, which is “over 95 km / hour immediately before the rollover of the bus”. Note that a space (blank) between “rollover” and “immediately before” in the shortened title AT11 is a half-width space. As described above, the generating apparatus 100 generates a summary of a shortened title or the like by appropriately using various symbols such as half-width and full-width, characters, and the like. Further, as shown in FIG. 9, the score corresponding to the probability that the word “Bus” appears at the top is “0.6”, and the score corresponding to the probability that the word “XX” appears at the top is “0”. .62 ”is larger, but since the score of the route of the shortened title AT11 with“ Bus ”at the head is higher than the score of the route of the shortened title with“ XX ”at the head, the generating apparatus 100 A shortened title AT11 is generated.

  As described above, the generation device 100 generates a character string including nine full-width characters and five half-width characters as the short title AT11. For example, when the half-width is “0.5” characters, the number of characters of the shortened title AT11 is “11.5”. As described above, the generation apparatus 100 generates the shortened title AT11 that satisfies 13 full-width characters. In the above example, the case where two half-width characters correspond to one full-width character and the width of each character does not vary is illustrated. The numerical value may be changed according to the character.

  Further, in the example of FIG. 1, the generation apparatus 100 generates a shortened title AT12 corresponding to “TSE market capitalization bubble period” corresponding to the title of the content CT12. In addition, the generation apparatus 100 generates a shortened title AT13 corresponding to the title of the content CT13, which is “Continuation of Company A surcharge July”. In addition, the generation apparatus 100 generates a shortened title AT14 corresponding to the title of the content CT14 and “determines the host country of the 20XX year”.

  Thereafter, the generation device 100 distributes the content including the shortened title to the terminal device 10 (step S14). Specifically, the generation device 100 distributes the contents CT11 to CT14 including each of the shortened titles AT11 to AT14 to the terminal device 10.

  And the terminal device 10 which received content CT11-CT14 displays content CT11-CT14 (step S15). In FIG. 1, the terminal device 10 displays the shortened titles AT11 to AT14 and the images IM11 to IM14 included in each of the contents CT11 to CT14 on a predetermined web page W11.

  As described above, the generation apparatus 100 generates a shortened title from a predetermined input based on the summary generation model M11 and user related information. In the above example, the generation device 100 generates a shortened title from the content title based on the summary generation model M11 and the user model. Accordingly, the generation device 100 generates a summary based on various user-related information such as user attribute information of the user, terminal information of the terminal device 10 used by the user, user context information, and the like. Appropriate summaries can be generated.

  Note that the generation apparatus 100 is not limited to the above, and may generate a shortened title using various information as appropriate. Further, in the above example, the generation apparatus 100 generates an abbreviated title so as to have one line, but the generation apparatus 100 may generate an abbreviated title (summary) extending over a plurality of lines. In addition, the generation apparatus 100 may generate a shortened title taking into account the width score of one character. For example, the generation apparatus 100 may generate a shortened title by taking into account a horizontal score in which one-byte alphabet “m” or “w” is one character and other half-width alphabets are 0.5 characters. For example, the generating apparatus 100 may calculate the score so that the score of a half-width alphabet that is 0.6 characters is higher than the score of a half-width alphabet that is 0.8 characters.

  In addition, the generation device 100 may change the generated short title according to the operation history of the user who uses the terminal device 10. For example, the generation apparatus 100 may use information related to a user operation on content distributed as a user operation history. For example, when the user's operation history includes information such as fast scrolling, the generation apparatus 100 may estimate the user's personality and generate a summary shorter than usual. Further, for example, when the user's operation history includes information such as slow scrolling, the generation apparatus 100 may estimate the user's personality as leisurely and generate a summary longer than usual.

  Further, for example, the generation apparatus 100 may use a voice conversation history as the user operation history. For example, the generation apparatus 100 may generate the summary so that the amount of information is larger than usual for the user who listens to the end in the voice dialogue. Further, the generation apparatus 100 may generate a summary so that the amount of information for a user who stops listening in the middle of a voice conversation is greater than the amount of information than usual. In addition, the generation apparatus 100 may use the above-described processing for changing the length by the user in a voice dialogue. For example, the generation apparatus 100 may use the above summary as output information of a voice dialogue, or may generate output information of a voice dialogue using a voice dialogue essential model.

  In addition, the generation device 100 may change the abbreviated title to be generated according to the attribute of the user who uses the terminal device 10. For example, the generation device 100 may generate a summary shorter than usual when the age of the user who uses the terminal device 10 is in his 60s or 70s. Further, for example, when the user who uses the terminal device 10 is a teenager or a twenties, the generation device 100 may generate a summary longer than usual.

  The generation apparatus 100 may perform generation processing using a model that generates a summary as described above. For example, when the user who uses the terminal device 10 is a professional job, the generation device 100 may generate a summary using a model that has a higher probability of including technical terms than usual. For example, when the user who uses the terminal device 10 is a housewife, the generation device 100 may generate a summary using a model that has a lower probability of including technical terms than usual. In addition, for example, when the job type of the user who uses the terminal device 10 is a profession related to the content of the content for which a shortened title is to be generated, the generating device 100 has a technical term score that is higher than that of a normally generated short title. A shortened title may be generated using a model that increases. For example, when the job type of the user who uses the terminal device 10 is a housewife, the generation device 100 may generate a shortened title using a model that reduces the technical term score.

  In addition, the generation device 100 may change the abbreviated title to be generated according to the context of the user who uses the terminal device 10. For example, when the context of the user using the terminal device 10 is a context corresponding to a busy state such as running, the generation device 100 may generate a summary shorter than usual. Further, for example, when the context of the user who uses the terminal device 10 is a context corresponding to work, the generation device 100 may generate a summary shorter than usual. In addition, for example, when the context of a user who uses the terminal device 10 is a context corresponding to a state with sufficient time such as being at home, the generation device 100 may generate a summary longer than usual. The generation apparatus 100 may perform generation processing using a model that generates a summary as described above.

  The generation apparatus 100 is not limited to the above, and may learn various models according to the purpose. For example, the generation apparatus 100 may include a document-to-document, a document-to-first paragraph, a document-to-first sentence, a document-to-title, a document-to-existing abbreviated title, a first-to-first paragraph, a first-to-first sentence, etc. The model may be learned in advance based on the correspondence relationship. The generation apparatus 100 may learn in advance using a general document. For example, the generating apparatus 100 may learn using various contents (information) such as general documents on the Web, not limited to the contents that are the target of the distribution service in the determination system 1. For example, the generation apparatus 100 may pre-learn (pretrain) the summary generation model based on the correspondence relationship such as a general document to a general document and a general document to a title.

  The generation apparatus 100 is not limited to the user-related information described above, and may use various information depending on the purpose. For example, the generation apparatus 100 may use information related to time during generation processing, information related to temperature, information related to weather, and the like as user-related information. The generation apparatus 100 may use information such as a CPU (Central Processing Unit) and a memory usage rate of the terminal apparatus 10 used by the user. Further, for example, the generation apparatus 100 clusters each piece of information such as user-related information, generates a summary (abbreviated title) in advance, stores it in the content storage unit 124, and responds to the requesting user when a distribution request is made. A shortened title may be distributed.

[2. Configuration of the generator
Next, the configuration of the generation apparatus 100 according to the embodiment will be described with reference to FIG. FIG. 2 is a diagram illustrating a configuration example of the generation apparatus 100 according to the embodiment. As illustrated in FIG. 2, the generation apparatus 100 includes a communication unit 110, a storage unit 120, and a control unit 130. The generation device 100 includes an input unit (for example, a keyboard and a mouse) that acquires various operations from an administrator of the generation device 100 and a display unit (for example, a liquid crystal display) for displaying various types of information. May be.

  The communication unit 110 is realized by, for example, a NIC (Network Interface Card). The communication unit 110 is connected to the network by wire or wirelessly, and transmits / receives information to / from the terminal device 10, the provider device, and the editor terminal.

(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 illustrated in FIG. 2, the storage unit 120 according to the embodiment includes an edit information storage unit 121, a model information storage unit 122, a user related information storage unit 123, and a content information storage unit 124.

(Edit information storage unit 121)
The editing information storage unit 121 according to the embodiment stores various types of information related to editing (hereinafter also referred to as “editing information”). For example, the editing information storage unit 121 stores various information related to a shortened title created by a predetermined editor and the title that is the basis of the shortened title. FIG. 3 shows an example of the editing information storage unit 121 according to the embodiment. The edit information storage unit 121 illustrated in FIG. 3 includes items such as “edit ID”, “original title”, and “short title”.

  “Edit ID” indicates identification information for identifying a combination of an original title and a shortened title. The “original title” indicates the title of the content that was used when a predetermined editor created a shortened title. “Abbreviated title” indicates a shortened title created from the corresponding original title.

  For example, in the example shown in FIG. 3, the edit information identified by the edit ID “ED11” indicates that the original title is “title A” and the shortened title is “shortened title A”. For example, in the example illustrated in FIG. 3, the editing information identified by the editing ID “ED12” indicates that the original title is “title B” and the shortened title is “shortened title B”.

  Note that the editing information storage unit 121 is not limited to the above, and may store various information according to the purpose. For example, the editing information storage unit 121 may store information related to an article of content that is a source when a predetermined editor creates a shortened title. In addition, the editing information storage unit 121 stores not only information on the abbreviated title created by a predetermined editor and the content that is the basis of the abbreviated title, but also information on the abbreviated title generated by the generation device 100 and the content that is the basis of the abbreviated title. May be. For example, the edit information storage unit 121 may store information for identifying a shortened title as a positive example and a shortened title as a negative example in association with each shortened title. That is, the editing information storage unit 121 may store a shortened title selected by the distribution destination user as a positive example and a shortened title not selected by the distribution destination user as a negative example. The generation apparatus 100 can learn a model for generating a summary (short title) more accurately by learning using edit information including a short title as a positive example and a short title as a negative example.

(Model information storage unit 122)
The model information storage unit 122 according to the embodiment stores information related to learning. For example, the model information storage unit 122 stores model information generated by the generation process. FIG. 4 is a diagram illustrating an example of a model information storage unit according to the embodiment. The model information storage unit 122 illustrated in FIG. 4 includes items such as “Feature 1” to “Feature 3” corresponding to each model M11 and the like as model information.

  For example, in the example illustrated in FIG. 4, the model information regarding the model M11 includes feature 1 weight “0.5”, feature 2 weight “−0.4”, feature 3 weight “0.2”, and the like. Indicates that For example, when the feature (feature value) of the model is expressed by an m-dimensional vector, the number of features is m, and the weights of the features 1 to m are stored.

  The model information storage unit 122 is not limited to the above, and may store various model information according to the purpose. For example, the model information storage unit 122 may store a model generated for each predetermined user classification. For example, the model information storage unit 122 may store a model generated for each user classification such as a user's sex and age. For example, the model information storage unit 122 may store a model generated for each classification related to the terminal device used by the user. For example, the model information storage unit 122 may store a model generated for each classification such as a model number or screen size of the terminal device.

(User related information storage unit 123)
The user related information storage unit 123 according to the embodiment stores various types of information related to the user. FIG. 5 shows an example of the user related information storage unit 123 according to the embodiment. The user related information storage unit 123 illustrated in FIG. 5 includes items such as “user ID”, “attribute information”, “terminal information”, and “context information”. The “attribute information” includes items such as “age”, “sex”, and “residence”. The “terminal information” includes items such as “model number”, “screen size”, and “font”. The “context information” includes items such as “current position”, “acceleration”, and “context”.

  “User ID” indicates identification information for identifying a user. For example, the user identified by the user ID “U1” corresponds to the user illustrated in the example of FIG. The “attribute information” includes various items related to user attributes. The “terminal information” includes various items related to the terminal device 10 used by the user. The “context information” includes various items related to the user context.

  “Age” indicates the age of the user identified by the user ID. The “age” may be a specific age of the user identified by the user ID, such as 35 years old. “Gender” indicates the gender of the user identified by the user ID. “Residence” indicates the residence of the user identified by the user ID.

  The “model number” indicates the model number of the terminal device 10 used by the user identified by the user ID. The “screen size” indicates the screen size of the terminal device 10 used by the user identified by the user ID. “Font” indicates information regarding the font set in the terminal device 10 used by the user identified by the user ID.

  “Current position” indicates the current position of the user identified by the user ID. “Acceleration” indicates the acceleration of the user identified by the user ID. “Context” indicates the context of the user identified by the user ID.

  For example, in the example shown in FIG. 5, the age of the user identified by the user ID “U1” is “30s”, the gender is “male”, and the residence is “region A”. Indicates.

  Note that the user-related information storage unit 123 is not limited to the above, and may store various information according to the purpose, such as demographic attribute information and psychographic attribute information other than the above. Further, the generation apparatus 100 may cause the user-related information storage unit 123 to store various types of information related to the user estimated based on the user information acquired from the terminal device 10.

(Content information storage unit 124)
The content information storage unit 124 according to the embodiment stores various types of information related to content (articles). FIG. 6 shows an example of the content information storage unit 124 according to the embodiment. The content information storage unit 124 illustrated in FIG. 6 includes items such as “content ID”, “article”, “title”, “abbreviated title”, and “image”.

  “Content ID” indicates identification information for identifying the content. “Article” indicates an article of content. “Title” indicates the title of the content. “Abbreviated title” indicates the abbreviated title of the content generated by the generation apparatus 100. “Image” indicates an image of content. FIG. 6 shows an example in which conceptual information such as “IM11” is stored in “image”, but actually image information or a file path name indicating the storage location is stored.

  For example, in the example shown in FIG. 6, the article identified by the content ID “CT11” is “X month Y day dawn, XX prefecture XX city prefectural road overturned. Indicates that the content exceeds 95km. The title identified by the content ID “CT11” indicates that the content is “X month Y day XX prefecture, bus rollover, immediately preceding speed, over 95 km / h”. Further, the abbreviated title (abbreviated title AT11 in FIG. 1) identified by the content ID “CT11” indicates that the content is “over 95 km / h just before the bus rollover”.

  The content information storage unit 124 is not limited to the above, and may store various types of information according to the purpose. For example, the content information storage unit 124 may store information regarding the date and time when the content was acquired and the date and time when the content was created. The content information storage unit 124 may store information on each provider and information on content categories.

(Control unit 130)
Returning to the description of FIG. 2, the control unit 130 is a controller, for example, various programs (generation programs) stored in a storage device inside the generation device 100 by a CPU, an MPU (Micro Processing Unit), or the like. This is realized by executing the RAM as a work area. The control unit 130 is a controller, and is realized by an integrated circuit such as an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA).

  As shown in FIG. 2, the control unit 130 includes an acquisition unit 131, a learning unit 132, a calculation unit 133, a generation unit 134, and a distribution unit 135, and functions and operations of information processing described below. Realize or execute. The internal configuration of the control unit 130 is not limited to the configuration illustrated in FIG. 2, and may be another configuration as long as the information processing described later is performed.

(Acquisition part 131)
The acquisition unit 131 acquires various types of information. For example, the acquisition unit 131 acquires information from an external information processing apparatus. For example, the acquisition unit 131 acquires information from the storage unit 120. The acquisition unit 131 acquires a content distribution request from the terminal device 10. For example, the acquisition unit 131 acquires content submission from the providing apparatus. For example, the acquisition unit 131 acquires a submission of content including an article and a title. For example, the acquiring unit 131 may acquire content submissions from a number of providing source devices. For example, the acquisition unit 131 stores the content submitted from the providing source device in the content information storage unit 124.

  In addition, the acquisition unit 131 acquires information in which the summary and the information that is the basis thereof are associated with each other. In FIG. 1, the acquisition unit 131 acquires a summary list EL11 in which a shortened title and an original title are associated with each other from an editor terminal used by a predetermined editor.

  In addition, the acquisition unit 131 acquires character information and user-related information that is information related to a user who is a provision destination. For example, the acquisition unit 131 acquires a title included in predetermined content as character information. For example, the acquisition unit 131 acquires an article included in predetermined content as character information.

  For example, the acquisition unit 131 acquires terminal information related to the terminal device 10 used by the user as user-related information. For example, the acquisition unit 131 acquires information related to an operation history in the user terminal device 10 as user-related information. For example, the acquisition unit 131 acquires user attribute information as user-related information. For example, the acquisition unit 131 acquires information related to the user context as user-related information.

  The acquisition unit 131 may acquire various models from an external information processing apparatus. For example, the acquisition unit 131 may acquire various models such as the summary generation model M11 and the user model from an external information processing apparatus. Note that when the acquisition unit 131 acquires a model used for summary generation from an external information processing device, the generation device 100 may not include the learning unit 132.

(Learning unit 132)
The learning unit 132 learns the weight of each feature. For example, the learning unit 132 generates a model by learning based on various information. For example, the learning unit 132 learns a model from information in which first character information and second character information that is summary information of the first character information are associated with each other. In FIG. 1, the learning unit 132 generates a summary generation model M11 from the summary list EL11 by learning using the summary list EL11.

  For example, the learning unit 132 acquires information shown in the summary list EL11 from the editing information storage unit 121, and learns the summary generation model M11. For example, the learning unit 132 learns the summary generation model M11 by using a combination of a shortened title created by a predetermined editor included in the summary list EL11 and the title that is the basis thereof. For example, the learning unit 132 learns the summary generation model M11 by appropriately using various conventional techniques such as a recurrent neural network and dependency analysis based on the original title and the abbreviated title included in the summary list EL11. For example, the learning unit 132 stores the summary generation model M11 generated by learning in the model information storage unit 122. For example, the learning unit 132 may learn the model using the following pseudo learning data. For example, a dependency analysis of "B line is delayed due to a person accident at station A" and other elements are related to "B line is delayed". Pseudo-learning data summarized by cutting a subtree and summarizing “Yamanote Line is delayed due to personal injury” is used. For example, such pseudo learning data can be generated by various conventional methods. The learning unit 132 can generate a highly accurate model by learning (pretraining) the model using the pseudo-learning data (pre-training) and then using the summary list EL11 and the like.

  The learning unit 132 may learn a model (user model) based on user-related information. For example, the learning unit 132 may learn a model corresponding to the function “U” in the right side of the equation (5). For example, when a predetermined shortened title is distributed to each user and the user selects the shortened title, the learning unit 132 learns the shortened title and the user related information of the user as a positive example. Thus, a user model may be generated.

  For example, the learning unit 132 may generate a user model corresponding to the terminal device 10 used by the user by learning based on the terminal information of the terminal device 10 used by the user. For example, the learning unit 132 may generate a user model according to information related to the user context by learning based on the user attribute information. For example, the learning unit 132 may generate a user model corresponding to information related to the user context by learning based on information related to the user context.

(Calculation unit 133)
The calculation unit 133 calculates a score used for generating a summary from character information. For example, the calculation unit 133 calculates the score at each position of each word using the summary generation model M11. For example, the calculation unit 133 calculates a score at each position of each word using the user model generated by the learning unit 132.

  For example, the calculation unit 133 calculates the score of each word at each position using the above equation (5). In the example illustrated in FIG. 1, the calculation unit 133 calculates a score that considers user-related information. For example, the calculation unit 133 calculates a score in consideration of the acquired user related information of the user U1. For example, the calculation unit 133 calculates a score in consideration of the user related information of the user U1 using the above formula (5).

(Generator 134)
The generation unit 134 generates a summary from the character information. For example, the generation unit 134 generates a summary from the character information based on the score calculated by the calculation unit 133. For example, the generation unit 134 generates summary information from the character information acquired by the acquisition unit 131 based on the user related information acquired by the acquisition unit 131.

  The generation unit 134 generates summary information from character information based on a model learned from information in which first character information and second character information that is summary information of the first character information are associated with each other. For example, the generation unit 134 generates summary information from character information based on a score calculated based on user-related information and a model. For example, the generation unit 134 generates summary information from character information based on a model learned from user-related information. For example, the generation unit 134 generates summary information from character information based on terminal information. For example, the generation unit 134 generates summary information from character information based on the operation history. For example, the generation unit 134 generates summary information from character information based on the attribute information. For example, the generation unit 134 generates summary information from character information based on information about the context. In the example of FIG. 1, the generation unit 134 generates a shortened title AT11 corresponding to the title of the content CT11. Further, the generation unit 134 generates shortened titles AT12 to AT14 corresponding to the titles of the contents CT12 to CT14.

  The generation unit 134 generates multiple summary information that further summarizes the generated summary information. For example, the generation unit 134 generates a title of an article as summary information from the article, and generates a shortened title as multiple summary information from the title. Details of these points will be described later.

(Distributor 135)
The distribution unit 135 distributes the shortened title generated by the generation unit 134. The distribution unit 135 distributes content including the shortened title to the terminal device 10. In the example of FIG. 1, the distribution unit 135 distributes content CT11 to CT14 including each of the shortened titles AT11 to AT14 to the terminal device 10.

[3. Model learning)
Next, model learning in the generation system 1 according to the embodiment will be described with reference to FIG. FIG. 7 is a flowchart illustrating an example of model learning according to the embodiment. Note that the model learned by the flowchart shown in FIG. 7 may be any model.

  As illustrated in FIG. 7, the acquisition unit 131 of the generation apparatus 100 acquires information used for learning (step S101). For example, the acquisition unit 131 acquires a summary list EL11 in which a shortened title and an original title are associated with each other from an editor terminal used by a predetermined editor. Thereafter, the learning unit 132 of the generation device 100 performs learning based on the acquired information (step S102). For example, the learning unit 132 generates a summary generation model M11 from the summary list EL11 by learning using the summary list EL11.

[4. (Generate abbreviated title)
Next, generation of a summary in the generation system 1 according to the embodiment will be described with reference to FIG. FIG. 8 is a flowchart illustrating an example of generation of a shortened title according to the embodiment.

  As illustrated in FIG. 8, the acquisition unit 131 of the generation apparatus 100 acquires a content distribution request (step S201). For example, the acquisition unit 131 acquires a content distribution request from the terminal device 10. Further, the acquisition unit 131 acquires user-related information of the request source (Step S202). For example, the acquisition unit 131 acquires user related information from the terminal device 10.

  Thereafter, the generation unit 134 of the generation device 100 determines an upper limit value of the number of characters of the shortened title based on the user related information (step S203). The generation unit 134 determines the upper limit value of the number of characters of the shortened title based on the screen size “SZ11” and the font “FT11” of the terminal device 10. For example, the generation unit 134 sets the upper limit of the number of characters of the shortened title to “13 full-width characters”. Further, the acquisition unit 131 acquires a model used for generating a shortened title (step S205). For example, the acquisition unit 131 acquires the summary generation model M11 from the model information storage unit 132.

  Thereafter, the generation unit 134 generates a shortened title based on the model and the upper limit value of the number of characters (step S204). For example, the generation unit 134 generates a shortened title AT11 corresponding to the title of the content CT11. In the example of FIG. 1, the generation unit 134 generates a shortened title AT11 corresponding to the title of the content CT11, which is “over 95 km / hour immediately before the bus rollover”.

[5. (Generate abbreviated title)
Next, generation of a shortened title in the generation system 1 according to the embodiment will be described with reference to FIG. FIG. 9 is a diagram illustrating an example of generation of a shortened title according to the embodiment. For example, the generation apparatus 100 uses a beam search technique. Note that the generation apparatus 100 may use various methods using a tree structure. For example, the generating apparatus 100 may use another approximation method, or may use an exact search method such as Viterbi or A *.

  In the example shown in FIG. 9, as shown in the input information IP11, the input “INPUT” is the title of the content CT11 “X month Y day XX prefecture bus rollover, immediately preceding speed, over 95 km / h”. In the example shown in FIG. 9, the upper limit “MAX” is “13” as indicated by the character number upper limit MX11. For example, the upper limit value “MAX” is “13 full-width characters”. In the tree structure ST11 in FIG. 9, a node located in the range of “W1” indicates a word that transitions after the first symbol S, that is, a node corresponding to the first word. The node located in the range of “W2” indicates a word that transitions next from the word corresponding to the node located in the range of “W1”, that is, the node corresponding to the second word. A node located in the range of “W3” indicates a word that transitions next from a word corresponding to the node located in the range of “W3”, that is, a node corresponding to the third word.

  In the example illustrated in FIG. 9, the generation device 100 calculates a score that considers user-related information. For example, the generating apparatus 100 calculates a score that takes into account the user-related information of the user U1 using the formula shown in the calculation formula MF11 in FIG. Note that the equation shown as the calculation equation MF11 in FIG. 9 corresponds to the above equation (5).

  For example, when the generation apparatus 100 generates an abbreviated title for the user-related information “U1” when the user is U1, the word “Bus” is first in the calculation formula SC11 in FIG. A score corresponding to the probability of appearance is calculated. The calculation formula SC11 in FIG. 9 corresponds to the above formula (7). In the example of FIG. 9, the generation device 100 calculates the score “0.62” as the highest score for the word “◯◯” using the calculation formula SC11. In the example of FIG. 9, the generation device 100 calculates the score of the word “bus” as “0.6”, which is the next higher than the score of the word “OO”, using the calculation formula SC11. In the example of FIG. 9, the generation apparatus 100 calculates the score of the word “rollover” as “0.58”, which is the next higher than the score of the word “bus”, using the calculation formula SC11. In the example of FIG. 9, the generation device 100 calculates the score of the word “NO” as “0.02” using the calculation formula SC11.

  Further, the generating apparatus 100 sorts the tree structure ST11 by the score (probability) calculated for each word using the above equation (5), and leaves only the beam width (n = 3) (step S205). -1). Specifically, the generation device 100 leaves only the nodes corresponding to the three words “◯◯”, “bus”, and “rollover” having the highest scores, and searches for the node ahead. The generating apparatus 100 repeats the search until the total number of characters exceeds MAX “13” (step S205-2). For example, the generation apparatus 100 uses the calculation formula MF12 in FIG. 9 to calculate the score corresponding to the probability of transition to each word next to the word (“OO”, “bus”, and “rollover”) of the parent node. Is calculated. As a result, the generation apparatus 100 generates the shortened title AT11 corresponding to the title of the content CT11, which is “over 95 km / hour just before the bus rollover”.

[6. Information used for learning the summary generation model)
In the above example, when the summary generation model M11 is generated, the summary list EL11 in which the shortened title is associated with the original title is used from the editor terminal used by the predetermined editor. Various information may be used for learning the summary generation model. For example, the generation device 100 may use the user related information for learning the summary generation model. This point will be described with reference to FIG. FIG. 10 is a diagram illustrating an example of model generation using user-related information according to the embodiment. The terminal device 10 shown in FIG. 10 shows a state in which the contents CT11 to CT14 distributed from the generation device 100 in step S15 of FIG. 1 are displayed.

  In the example of FIG. 10, the user U1 selects the content CT11 including the shortened title AT11 (step S21). For example, the user U1 selects the content CT11 by touching the portion where the shortened title AT11 is displayed with the finger F1. Further, for example, when the user U1 selects the shortened title AT11 on the terminal device 10, the display of the terminal device 10 transitions, and the title and article of the content CT11 are displayed.

  Thereafter, the terminal device 10 transmits the behavior information of the user U1 to the generation device 100 (Step S22). For example, the terminal device 10 transmits information related to the operation history of the user U1 to the generation device 100. In the example of FIG. 10, the terminal device 10 transmits information indicating that the user U1 has selected the content CT11 including the shortened title AT11 to the generation device 100.

  The generating apparatus 100 that has acquired the action information of the user U1 performs learning by using the combination of the shortened title AT11 of the content CT11 and the title that is the original as the positive example RL11. For example, the generation device 100 generates the summary generation model M111 by updating the summary generation model M11 using the user-related information of the positive example RL11 and the user U1 (step S23). Note that the generation apparatus 100 may newly generate a summary generation model using the user-related information of the positive example RL11 and the user U1. As described above, the generation apparatus 100 may learn the summary generation model using the user related information.

[7. (Multiple summary)
In the above example, the case where the summary is generated from the character information is shown, but the summary may be further generated using the generated summary. For example, the generating apparatus 100 may generate a multiple summary that further summarizes the generated summary. That is, the generation apparatus 100 may generate a desired summary by performing multiple summaries from one character information. This point will be described with reference to FIG. FIG. 11 is a diagram illustrating an example of generation of a two-stage summary according to the embodiment.

  In the example of FIG. 11, the article CR11 includes contents such as “Decimal day of X month Y, XX prefecture XX city prefectural road overturned Z ... the speed just before exceeds 95 km / h ..." . The generation apparatus 100 generates the title of the article CR11 by inputting the article CR11 to the summary generation model M31 that generates a title from the article (step S31). For example, the generation device 100 generates the title TL11 from the article CR11 using the summary generation model M31.

  In the example of FIG. 11, the title TL11 is a title with the content “X month Y day XX prefecture, bus rollover, previous speed, over 95 km / h”. The generation apparatus 100 generates the shortened title of the title TL11 by inputting the title TL11 to the summary generation model M11 that generates a shortened title from the title (step S32). For example, the generating apparatus 100 generates a shortened title AT11 from the title TL11 using the summary generation model M11. In this way, the generation apparatus 100 can generate a desired summary by summarizing one character information multiple times using a summary generation model. In the above example, the generation device 100 generates a summary multiple times using a plurality of summary generation models. However, the generation device 100 generates a summary multiple times using a single summary generation model. May be. For example, the generating apparatus 100 learns (pretrains) the model M using the first data of the combination of the text and the title, and uses the weight as an initial value to set the second combination of the title and the shortened title. Learning (fine tuning) may be performed using data. In this case, when the generation apparatus 100 has a small amount of second data of a combination of a title and its abbreviated title and a large amount of first data of a combination of a text and its title (for example, 10 times or more), Learning can be performed on the basis of the first data 1 although there are no combinations. Further, the generation apparatus 100 may generate a shortened title directly from an article by using a summary generation model that generates a shortened title from the article.

[8. Variations in shortened titles according to user-related information)
In addition, the summary which the production | generation apparatus 100 produces | generates changes based on the user relevant information of the user who becomes a delivery destination. For example, the generation device 100 may change the summary to be generated according to the terminal device 10 used by the user. This point will be described with reference to FIG. FIG. 12 is a diagram illustrating an example of generation of a shortened title according to the terminal device according to the embodiment. In the example of FIG. 12, it is assumed that the font settings in the terminal devices 10 and 11 are the same.

  In the example illustrated in FIG. 12, a case will be described in which the generation device 100 generates a shortened title AT11 corresponding to the content CT11 for each terminal device 10 and 11. In the example of FIG. 12, the generation device 100 generates a shortened title AT11 for the terminal device 10 (step S41). Specifically, the generation device 100 generates the shortened title AT11 corresponding to the title of the content CT11, which is “over 95 km / hour immediately before the bus rollover”, so that it can be displayed in the display area AR11. Note that the terminal device 10 in FIG. 12 corresponds to the terminal device 10 in FIG. 1, and the horizontal width of the terminal device 10 is WD11.

  In the example of FIG. 12, the generation device 100 generates a shortened title AT41 for the terminal device 11 (step S42). Here, in the example of FIG. 12, the lateral width WD41 of the terminal device 11 is longer than the lateral width WD11 of the terminal device 10. For this reason, the display area AR41 in which the shortened title of the content CT11 is displayed in the terminal device 11 is longer than the display area AR11 in which the shortened title of the content CT11 is displayed in the terminal device 10. Specifically, the horizontal length L41 of the display area AR41 in the terminal device 11 is longer than the horizontal length L11 of the display area AR11 in the terminal device 10. That is, more characters can be displayed in the display area AR41 than in the display area AR11. Therefore, the generation apparatus 100 generates a shortened title AT41 having a larger number of characters than the shortened title AT11.

  Specifically, the generation device 100 generates a shortened title AT41 corresponding to the title of the content CT11, which is “○○ exceeds 95 km / h just before the bus rollover”, so that the generation apparatus 100 can display the display area AR41. In this case, the generation apparatus 100 can add a word indicating a specific place “OO” to the abbreviated title AT41 delivered to the terminal apparatus 11 as compared to the abbreviated title AT11. As described above, the generation device 100 can generate an appropriate summary according to the user-related information. Note that the generation apparatus 100 may generate a summary using not only user-related information stored in the user-related information storage unit 123 but also various user-related information. For example, the generation apparatus 100 may dynamically change the summary by acquiring in real time an enlargement factor or font that has been temporarily changed by the user.

[9. effect〕
As described above, the generation device 100 according to the embodiment includes the acquisition unit 131 and the generation unit 134. The acquisition unit 131 acquires character information and user-related information that is information related to a user who is a providing destination. The generation unit 134 generates summary information from the character information acquired by the acquisition unit 131 based on the user-related information acquired by the acquisition unit 131.

  Thereby, the production | generation apparatus 100 which concerns on embodiment can produce | generate an appropriate summary according to a user by considering the user relevant information of the user who becomes a provision destination, and producing | generating a summary from character information. As described above, the generation apparatus 100 generates a summary based on various user-related information such as user attribute information of the user, terminal information of the terminal apparatus 10 used by the user, user context information, and the like. Can generate an appropriate summary.

  Further, in the generation device 100 according to the embodiment, the generation unit 134 includes first character information (in the embodiment, “title”; the same applies hereinafter) and second character information that is summary information of the first character information (the embodiment). In the example, summary information is generated from character information based on a model learned from information associated with “abbreviated title” (the same applies hereinafter).

  Thereby, the generating apparatus 100 according to the embodiment generates a summary using a model learned from information in which the first character information and the second character information that is summary information of the first character information are associated with each other. Thus, an appropriate summary can be generated according to the user.

  In the generation device 100 according to the embodiment, the generation unit 134 generates summary information from character information based on a score calculated based on user-related information and a model.

  Thereby, the production | generation apparatus 100 which concerns on embodiment can produce | generate an appropriate summary according to a user by producing | generating a summary based on the score calculated based on a user relevant information and a model.

  Further, in the generation device 100 according to the embodiment, the generation unit 134 generates summary information from character information based on a model learned from user-related information.

  Thereby, the production | generation apparatus 100 which concerns on embodiment can produce | generate an appropriate summary according to a user by producing | generating a summary based on the model learned by the user relevant information.

  In the generation device 100 according to the embodiment, the acquisition unit 131 acquires terminal information related to the terminal device 10 used by the user as user-related information. The generation unit 134 generates summary information from the character information based on the terminal information.

  Thereby, the production | generation apparatus 100 which concerns on embodiment can produce | generate an appropriate summary according to a user by producing | generating a summary using the terminal information regarding the terminal device 10 which a user uses.

  Further, in the generation device 100 according to the embodiment, the acquisition unit 131 acquires information related to an operation history of the user terminal device 10 as user-related information. The generation unit 134 generates summary information from the character information based on the operation history.

  Thereby, the production | generation apparatus 100 which concerns on embodiment can produce | generate an appropriate summary according to a user by producing | generating a summary using the information regarding the operation history in a user's terminal device 10. FIG.

  In the generation device 100 according to the embodiment, the acquisition unit 131 acquires user attribute information as user-related information. The generation unit 134 generates summary information from the character information based on the attribute information.

  Thereby, the production | generation apparatus 100 which concerns on embodiment can produce | generate an appropriate summary according to a user by producing | generating a summary using a user's attribute information.

  In the generation device 100 according to the embodiment, the acquisition unit 131 acquires information related to the user context as user-related information. The generation unit 134 generates summary information from the character information based on the information regarding the context.

  Thereby, the production | generation apparatus 100 which concerns on embodiment can produce | generate an appropriate summary according to a user by producing | generating a summary using the information regarding a user's context.

  In the generation device 100 according to the embodiment, the generation unit 134 generates multiple summary information that further summarizes the generated summary information.

  Thus, the generation apparatus 100 according to the embodiment can generate a desired summary according to the user by generating multiple summary information that further summarizes the generated summary information.

  In the generation device 100 according to the embodiment, the acquisition unit 131 acquires an article included in predetermined content as character information. The generation unit 134 generates an article title from the article as summary information, and generates a shortened title from the title as multiple summary information.

  Thus, the generation apparatus 100 according to the embodiment generates an article title as summary information from the article, and generates an abbreviated title as multiple summary information from the title, thereby generating an appropriate shortened title according to the user. be able to.

[10. Hardware configuration)
The generation apparatus 100 according to the above-described embodiment is realized by a computer 1000 having a configuration as illustrated in FIG. FIG. 13 is a hardware configuration diagram illustrating an example of a computer that realizes the function of the generation apparatus. The computer 1000 includes a CPU 1100, RAM 1200, ROM 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 a program stored in the ROM 1300 or the HDD 1400 and controls each unit. The ROM 1300 stores a boot program executed by the CPU 1100 when the computer 1000 is started up, a program depending 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. The communication interface 1500 receives data from other devices via the network N and sends the data to the CPU 1100, and transmits data determined by the CPU 1100 to other devices via the network N.

  The CPU 1100 controls an output device such as a display and a printer and an input device such as a keyboard and a mouse via the input / output interface 1600. The CPU 1100 acquires data from the input device via the input / output interface 1600. Further, the CPU 1100 outputs the determined data to the output device via the input / output interface 1600.

  The media interface 1700 reads a program or data stored in the recording medium 1800 and provides it to the CPU 1100 via the RAM 1200. The CPU 1100 loads such a program from the recording medium 1800 onto the RAM 1200 via the 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 PD (Phase change rewritable disk), 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 generation apparatus 100 according to the embodiment, the CPU 1100 of the computer 1000 implements the function of the control unit 130 by executing a program loaded on the RAM 1200. The CPU 1100 of the computer 1000 reads these programs from the recording medium 1800 and executes them. However, as another example, these programs may be acquired from other devices via the network N.

  As described above, some of the embodiments of the present application have been described in detail with reference to the drawings. It is possible to implement the present invention in other forms with improvements.

[11. Others]
In addition, among the processes described in the above embodiment, all or part of the processes described as being automatically performed can be performed manually, or the processes described as being performed manually can be performed. All or a part 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 document and drawings can be arbitrarily changed unless otherwise specified. For example, the various types of information illustrated in each drawing is not limited to the illustrated information.

  Further, each component of each illustrated apparatus is functionally conceptual, and does not necessarily need to be physically configured as illustrated. In other words, the specific form of distribution / integration of each device is not limited to that shown in the figure, and all or a part thereof may be functionally or physically distributed or arbitrarily distributed in arbitrary units according to various loads or usage conditions. Can be integrated and configured.

  Further, the above-described embodiments can be appropriately combined within a range in which processing contents are not contradictory.

  In addition, the “section (module, unit)” described above can be read as “means” or “circuit”. For example, the acquisition unit can be read as acquisition means or an acquisition circuit.

DESCRIPTION OF SYMBOLS 1 Generation system 100 Generation apparatus 121 Edit information storage part 122 Model information storage part 123 User related information storage part 124 Content information storage part 130 Control part 131 Acquisition part 132 Learning part 133 Calculation part 134 Generation part 135 Distribution part

In addition, the generation device 100 may change the generated short title according to the operation history of the user who uses the terminal device 10. For example, the generation apparatus 100 may use information related to a user operation on content distributed as a user operation history. For example, generator 100, if the scroll during operation history of the user include information such as have fast, estimates the characteristics of the user and impatient, may generate a short summary than usual. Further, for example, when the user's operation history includes information such as slow scrolling, the generation apparatus 100 may estimate the user's personality as leisurely and generate a summary longer than usual.

Claims (12)

An acquisition unit that acquires character information and user-related information that is information about a user who is a providing destination;
Based on the user related information acquired by the acquisition unit, a generation unit that generates summary information from the character information acquired by the acquisition unit;
A generating apparatus comprising: The generator is
Summarization information is generated from the character information based on a model learned from information in which first character information and second character information that is summary information of the first character information are associated with each other. The generation device according to claim 1. The generator is
The generating apparatus according to claim 2, wherein the summary information is generated from the character information based on a score calculated based on the user-related information and the model. The generator is
4. The generation apparatus according to claim 2, wherein summary information is generated from the character information based on the model learned from the user-related information. 5. The acquisition unit
Obtaining terminal information related to the terminal device used by the user as the user-related information;
The generator is
The generation apparatus according to claim 1, wherein the summary information is generated from the character information based on the terminal information. The acquisition unit
Obtaining information related to operation history in the terminal device of the user as the user-related information;
The generator is
The generation apparatus according to claim 1, wherein the summary information is generated from the character information based on the operation history. The acquisition unit
Obtaining the user attribute information as the user-related information;
The generator is
The generation apparatus according to any one of claims 1 to 6, wherein the summary information is generated from the character information based on the attribute information. The acquisition unit
Obtaining information about the user's context as the user-related information;
The generator is
The generation apparatus according to claim 1, wherein the summary information is generated from the character information based on information related to the context. The generator is
The generation apparatus according to claim 1, wherein multiple summary information is generated by further summarizing the generated summary information. The acquisition unit
An article included in predetermined content is acquired as the character information,
The generator is
The generation apparatus according to claim 9, wherein a title of the article is generated as the summary information from the article, and a shortened title is generated as the multiple summary information from the title. A generation method executed by a computer,
An acquisition step of acquiring character information and user-related information that is information about a user who is a providing destination;
Based on the user-related information acquired by the acquisition step, a generation step of generating summary information from the character information acquired by the acquisition step;
A generation method comprising: An acquisition procedure for acquiring character information and user-related information that is information about a user who is a providing destination;
Based on the user related information acquired by the acquisition procedure, a generation procedure for generating summary information from the character information acquired by the acquisition procedure;
A program for causing a computer to execute.

Images