JP2021103542A - Information providing device, information providing method, and program - Google Patents

Abstract

To provide an information providing device capable of highly accurately recommending content assumed by a user as having high interest, an information providing method, and a program.SOLUTION: An information providing device comprises a derivation unit and a determination unit. The derivation unit derives features of a user on the basis of features of content browsed by the user. The determination unit determines content recommended to the user from among plural pieces of content to be distributed on the basis of the features of the user derived by the derivation unit and features of each of the plural pieces of content to be distributed. The derivation unit derives features of the users on the basis of a change in the number of views according to the time course of each of content browsed by the user.SELECTED DRAWING: Figure 2

Description

The present invention relates to an information providing device, an information providing method, and a program.

Technology for distributing content such as news articles and advertising images via the Internet is known. In relation to this, a technique is known in which the importance of content is estimated, and the higher the importance of the content, the higher the priority when recommending to the user (see, for example, Patent Document 1).

JP-A-2017-59057

However, in the conventional technology, the content presumed to be of high importance may be content that is of little interest to the user. As a result, there are cases where content that is expected to be of low interest to the user is preferentially recommended.

The present invention has been made in view of the above problems, and provides an information providing device, an information providing method, and a program capable of accurately recommending contents that are expected to be of great interest to the user. I am aiming.

One aspect of the present invention is a derivation unit that derives the characteristics of the user based on the characteristics of the content viewed by the user, the characteristics of the user derived by the derivation unit, and a plurality of contents to be distributed. A determination unit for determining content to be recommended to the user from a plurality of contents to be distributed based on the characteristics of each is provided, and the derivation unit includes a time lapse of each content viewed by the user. This is an information providing device that derives the characteristics of the user based on the change in the number of times of browsing according to the above.

According to one aspect of the present invention, it is possible to accurately recommend content that is expected to be of great interest to the user.

It is a figure which shows an example of the information providing system 1 including the information providing apparatus 100 in embodiment. It is a figure which shows an example of the structure of the information providing apparatus 100 in embodiment. It is a figure which shows an example of the content information 132. It is a figure which shows an example of the content vector information 134. It is a figure which shows an example of the user information 136. It is a figure which shows an example of the action log 138. It is a figure which shows an example of the user vector information 140. It is a flowchart which shows the flow of a series of processing by an information providing apparatus 100. It is a flowchart which shows the flow of a series of processing by an information providing apparatus 100. It is a figure which shows an example of the relationship between the parameter b _pv and the time change of the number of times of viewing a content. It is a figure which shows typically the content of the process at the time of determining the parameter of a learner. It is a figure which shows typically the content of the process at the time of determining the parameter of a learner. It is a figure which shows an example of the hardware composition of the information providing apparatus 100 of an embodiment.

Hereinafter, an information providing device, an information providing method, and a program to which the present invention is applied will be described with reference to the drawings.

[Overview]
The information providing device is realized by one or more processors. The information providing device basically derives the user's characteristics based on the characteristics of the content viewed by the user, and based on the derived user's characteristics and the characteristics of each of the plurality of contents to be distributed. , Determine the content to be recommended to the user from a plurality of contents to be distributed. Further, when the information providing device derives the characteristics of the user, among the characteristics of one or more contents viewed by the user, the characteristics of the content in which the change in the number of views with the passage of time is small becomes the characteristics of the user. To reflect. This makes it difficult to reflect the characteristics of the gossip article in the characteristics of the user even when the target user browses the content such as the gossip article temporarily viewed by an unspecified number of users. be able to. As a result, it becomes difficult to recommend content that is instantaneously viewed by an unspecified number of users, such as gossip articles, and it is possible to accurately recommend content that is expected to be of great interest to users.

The content in the present embodiment is, for example, an article posted on a blog, a website, or the like, and is content including text. Further, the content may be, for example, text, a still image, a moving image, an audio, or the like posted as an advertisement on a medium such as a website.

[overall structure]
FIG. 1 is a diagram showing an example of an information providing system 1 including an information providing device 100 in the embodiment. The information providing system 1 in the embodiment includes, for example, one or more terminal devices 10 and an information providing device 100. These devices are connected via the network NW.

Each device shown in FIG. 1 transmits and receives various information via the network NW. The network NW includes, for example, the Internet, a WAN (Wide Area Network), a LAN (Local Area Network), a provider terminal, a wireless communication network, a wireless base station, a dedicated line, and the like. It should be noted that not all combinations of the devices shown in FIG. 1 need to be able to communicate with each other, and the network NW may include a local network in part.

The terminal device 10 is a terminal device including an input device, a display device, a communication device, a storage device, and an arithmetic device, such as a mobile phone such as a smartphone, a tablet terminal, and various personal computers. The communication device includes a network card such as a NIC (Network Interface Card), a wireless communication module, and the like. In the terminal device 10, a UA (User Agent) such as a web browser or an application program is activated, and a request according to the content input by the user is transmitted to the information providing device 100. Further, the terminal device 10 in which the UA is activated causes the display device to display various images based on the information acquired from the information providing device 100.

The information providing device 100 is, for example, a web server that provides a web page to the terminal device 10 in response to a request from a web browser. The above-mentioned contents are posted on the web page. Further, the information providing device 100 may be an application server that provides contents such as images and sounds to the terminal device 10 in response to a request from the application program. For example, the information providing device 100 is a user who uses a terminal device 10 via a website or an application to perform a service (hereinafter referred to as a search service) for searching a desired content from a certain database, such as a document search or an image search. To provide.

[Configuration of information providing device]
FIG. 2 is a diagram showing an example of the configuration of the information providing device 100 in the embodiment. As shown in the figure, the information providing device 100 includes, for example, a communication unit 102, a control unit 110, and a storage unit 130.

The communication unit 102 includes, for example, a communication interface such as a NIC. The communication unit 102 communicates with the terminal device 10 via the network NW and acquires various information from the terminal device 10. For example, the communication unit 102 acquires login information from the terminal device 10. The login information is information including a user ID (account ID) and a password associated with an account for using the search service.

The control unit 110 includes, for example, a content vector generation unit 112, a user vector generation unit 114, a recommendation content determination unit 116, a content distribution unit 118, an information update unit 120, and a learning processing unit 122. The user vector generation unit 114 is an example of the “derivation unit”, and the recommendation content determination unit 116 is an example of the “determination unit”.

These components are realized, for example, by a processor such as a CPU (Central Processing Unit) executing a program stored in the storage unit 130. In addition, some or all of the components of the control unit 110 are hardware such as LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), or GPU (Graphics Processing Unit). It may be realized by (circuit unit; circuitry), or it may be realized by the cooperation of software and hardware.

The storage unit 130 is realized by, for example, a storage device such as an HDD (Hard Disc Drive), a flash memory, an EEPROM (Electrically Erasable Programmable Read Only Memory), a ROM (Read Only Memory), and a RAM (Random Access Memory). In addition to various programs such as firmware and application programs, the storage unit 130 stores content information 132, content vector information 134, user information 136, action log 138, user vector information 140, learner information 142, and the like.

The content information 132 is information including various contents. For example, when the content is a news article, the content information 132 may include text information, images, audio, and the like included in the news article, and information such as a news article distributor and author.

FIG. 3 is a diagram showing an example of the content information 132. As in the illustrated example, the content information 132, the content ID for identifying each content may be information Total Views are associated for each predetermined period T _i. Predetermined period T _i is, for example, minute, hour, day, week, month, may be any of the time periods, such as year. Further, the content information 132 may be information including various contents. For example, when the content is a news article, the text information included in the news article, the content such as an image, and the information such as the distribution source and the author of the news article. May include.

FIG. 4 is a diagram showing an example of the content vector information 134. The content vector information 134 is information in which the features of each content included in the content information 132 are vectorized. For example, the content vector is associated with each content ID. ..

FIG. 5 is a diagram showing an example of user information 136. The user information 136 is, for example, information in which a user ID (account ID) for identifying each user is associated with a password, a user attribute, and the like that are collated at the time of login. User attributes include, for example, demographic attributes (demographic attributes) possessed by the user, such as gender, age, area of residence, income, occupation, and educational background. These demographic attributes shall be pre-registered, for example, when the account is created. Further, in the user attribute, for example, the type of the terminal device 10 indicating whether the terminal device 10 used by the user at the time of login is a mobile phone such as a smartphone, a tablet terminal, or a personal computer. Or, various information such as the day and time of login may be included as attributes.

FIG. 6 is a diagram showing an example of the action log 138. The action log 138 is information based on the browsing history of the content by the user. For example, for each user ID, the content ID of the content browsed by the user and the number of times the content is browsed in a predetermined period are associated with each other. Is. Predetermined period is, for example, a period until the user several periods T _i or several tens period T _i from the log has passed. Further, the predetermined period may be, for example, a period from the user logging in to logging out.

FIG. 7 is a diagram showing an example of the user vector information 140. The user vector information 140 is information in which the characteristics of each user indicated by the user information 136 are vectorized. For example, the user ID is associated with a user vector in which the user is vectorized.

The learning device information 142 is information that defines a learning device (learning model). The learner may be realized by, for example, a recurrent neural network (RNN) of a type such as a GRU (Gated Recurrent Unit) or an LSTM (Long short-term memory). When a learner is realized by these recursive neural networks, the learner information 142 includes an input layer, one or more intermediate layers (hidden layers), and an output layer, respectively, which constitute the recursive neural network. It includes various information such as connection information on how the neurons (units) included in the are connected to each other, and the connection coefficient given to the data input / output between the connected neurons. The connection information includes, for example, information such as the number of neurons contained in each layer, information that specifies the neuron to which each neuron is connected, an activation function that realizes each neuron, and a gate provided between neurons in the hidden layer. Including. The activation function that realizes the neuron may be, for example, a rectified linear function (ReLU function), a sigmoid function, a step function, or other function. The gate selectively passes data transmitted between neurons, for example, depending on the value returned by the activation function (eg 1 or 0). The coupling coefficient is a parameter of the activation function. For example, in a hidden layer of a neural network, when data is output from a neuron in a certain layer to a neuron in a deeper layer, the weight given to the output data is given. including. Further, the coupling coefficient may include a bias component peculiar to each layer.

The content vector generation unit 112 vectorizes the features of each content included in the content information 132 and generates a content vector. For example, when the content is a news article, the content vector generation unit 112 vectorizes the news article based on the learning model of Countuous Bag-of-Words and / or Skip-gram, using the news article as a corpus. For example, the content vector generation unit 112 divides a news article into a plurality of morphemes, and among the plurality of morphemes, a specific part of speech (noun, etc.) is a learning model of both Countuous Bag-of-Words and Skip-gram. By inputting as a word into a neural network (for example, word2vec) to which is applied, the word vector of each word output from the neural network is acquired. Then, the content vector generation unit 112 uses a value obtained by normalizing the sum of a plurality of word vectors corresponding to each word (for example, dividing by the number of words) as a vector of a news article (an example of a content vector) that is a word extraction source. Derived as. That is, the content vector generation unit 112 generates a vector of news articles distributed and expressed by a plurality of word vectors. The vector of this news article is a multidimensional vector in which each of a plurality of word vectors is a dimension (base vector).

Further, the content vector generation unit 112 may vectorize each news article by using a learning model composed of Bag-of-Words alone or another learning model such as Global Vector Prediction (GloVe). Further, when the content is a still image or a moving image, the content vector generation unit 112 may vectorize the content by using, for example, a convolutional neural network.

The content vector generation unit 112 associates the generated content vector with the content ID of the content that is the source of the content vector, and stores this as the content vector information 134 in the storage unit 130. The process of vectorizing the features of the content may be realized by requesting the external device to perform the process instead of the content vector generation unit 112 performing the process by itself.

The user vector generation unit 114 vectorizes the characteristics of each user included in the user information 136 based on the content vector information 134 and the action log 138, and generates a user vector. For example, the user vector generation unit 114 refers to the learner information 142 to generate (construct) a recurrent neural network which is an example of the learner. On the other hand, the user vector generation unit 114 refers to the action log 138, acquires the content ID of the content viewed by the user from the storage unit 130 for each user ID, and then refers to and acquires the content vector information 134. The content vector corresponding to the content ID is acquired from the storage unit 130. Then, the user vector generation unit 114 inputs each content vector corresponding to the content ID of the content viewed by the user to the input layer of the generated recurrent neural network, so that the user can use the output layer of the recurrent neural network. Get a user vector whose features are vectorized. That is, the user vector generation unit 114 generates a user vector distributed and expressed by a plurality of content vectors. This user vector is a multidimensional vector in which each of a plurality of content vectors is a dimension (base vector). The user vector is an example of "user characteristics".

In the case of the example of FIG. 6, the action log 138 indicates that the user whose user ID is "USER_2" has browsed the content whose content IDs are "CONT_6" and "CONT_14". In such a case, the user vector generation unit 114 refers to the content vector information 134, and corresponds to the content vector V (CONT_6) corresponding to the content having the content ID "CONT_6" and the content having the content ID "CONT_14". At least the content vector V (CONT_14) is acquired, and a vector set containing at least these content vectors is input to the recursive neural network. In the recurrent neural network, each input content vector is given a coupling coefficient (weight) in the hidden layer and averaged (for example, weighted averaging) up to the output layer. Then, the user vector generation unit 114 derives one vector obtained by averaging a plurality of content vectors by the recurrent neural network as a user vector of a user whose user ID is "USER_2".

The user vector generation unit 114 associates the generated user vector with the user ID of the user who is the source of the user vector, and stores this as user vector information 140 in the storage unit 130.

The recommendation content determination unit 116 is generated by the content vector generation unit 112 and the user vector in which the target user is vectorized (hereinafter referred to as the target user vector) among the plurality of user vectors generated by the user vector generation unit 114. A score (index value) based on the similarity with each of the one or more content vectors is derived, and the content recommended to be viewed by the target user (hereinafter referred to as recommended content) is determined according to the derived score. To do. The target user is, for example, a user who has entered a query in a search window or the like when searching for desired information using a search service after logging in.

For example, the recommendation content determination unit 116 determines the content corresponding to one content vector having the highest score as the recommendation content. Further, the recommendation content determination unit 116 may determine a set of contents corresponding to each of the upper predetermined number of content vectors having a large score as the recommendation content, or the recommendation content determination unit 116 of all the content vectors having a score equal to or higher than the threshold value. The set of contents corresponding to each may be determined as the recommended content.

The content distribution unit 118 uses the communication unit 102 to distribute one or more contents determined as recommended contents by the recommendation content determination unit 116 to the terminal device 10 used for inputting a query by the target user. ..

For example, when the recommended content distributed by the content distribution unit 118 is viewed by the target user, the information update unit 120 may newly add the content ID of the recommended content to the user ID of the target user in the action log 138. The action log 138 is updated by replacing it with the content ID of the content whose browsing time is old. At this time, the user vector generation unit 114 may newly generate the target user vector based on the updated action log 138 when the target user vector does not exist in the user vector information 140. Further, the user vector generation unit 114 may update the target user vector based on the updated action log 138 when the target user vector already exists in the user vector information 140.

In the action log 138, the information update unit 120 browses any content within a predetermined period when the content ID exists in the user ID of the target user, that is, when the target user vector already exists. If not, in the action log 138 of the target user, all the content IDs associated with the user IDs may be deleted and the browsing history may be initialized. In this case, the user vector generation unit 114 may erase the already generated target user vector, for example.

The learning processing unit 122 learns the learning device defined by the learning device information 142 based on the action log 138. For example, if the learner is recurrent neural network, the learning processing unit 122, based on the browsing history of the contents of each user until one period T _i elapses, the coupling coefficient of the recursive neural network (activity Machine learning of recurrent function parameters). The details of the learning method of the learner will be described later.

[Processing flow during operation]
Hereinafter, a series of processing flows during operation of the information providing device 100 will be described with reference to a flowchart. The operation time is a state in which the learning device learned to some extent by the learning processing unit 122 is used. FIG. 8 is a flowchart showing a flow of a series of processes by the information providing device 100. The processing of this flowchart is performed, for example, when a query for searching the content is further acquired from the terminal device 10 for which the login information has been acquired by the communication unit 102.

First, the user vector generation unit 114 refers to the content vector information 134 and the action log 138, and acquires the content vector vector of the content viewed by the target user from the storage unit 130 (S100).

Next, the user vector generation unit 114 inputs the acquired plurality of content vectors into the learner indicated by the learner information 142 to generate the target user vector in which the characteristics of the target user are distributed and expressed by the plurality of content vectors. Generate (S102).

Next, the recommendation content determination unit 116 includes a target user vector generated by the user vector generation unit 114 and one or more content vectors generated by the content vector generation unit 112 based on the mathematical formula (1). Based on the similarity of, the score to be referred to when determining the recommended content is derived (S102).

In the formula (1), u represents a target user among a plurality of users whose action log 138 includes a user ID, and a is selected as a score derivation target among a plurality of contents for which a content vector is generated. It represents one content, and t represents the time when the content was posted in the search service (in the case of a news article, the time when the content was submitted). R (u, a, t) represents a _{score, V cont} (a) represents a content vector corresponding to the content _{a, V user} (u), expressed the target user vector corresponding to the target user u There is. T represents transpose. pv (a, t) represents the index value based on the number of views per predetermined period T _i of the content a in an arbitrary time t (for example, the logarithm of the Views). b _pv represents a bias component that is multiplied by pv (a, t). Since pv (a, t) does not include u in the variable, the second term (b _pv · pv (a, t)) refers to the first term when viewed from the first term on the right side of the mathematical formula (1). Treated as a constant term to add or subtract. The parameter b _pv is also learned by the learning processing unit 122 when the coupling coefficient of the recurrent neural network is learned. The parameter b _pv is learned so that the smaller the change in the number of views (for example, the time derivative value of the number of views) of the content a selected as the score derivation target with the passage of time, the larger the value.

For example, the recommended content determining unit 116, the first term on the right-hand side of Equation (1), for each content vector _{V cont} (a), derives the cosine similarity between the target user vector _{V user} (u). On the other hand, the recommendation content determination unit 116 derives the product of the _{index value pv (a, t) and the bias value b pv} based on the number of views of the content a at time t as the second term on the right side of the mathematical formula (1). To do. Then, the recommendation content determination unit 116 derives the sum of these as a score R (u, a, t). In Equation (1), which is the first term of the right side ^{_{V cont (a) T · V}} user (u) is the value in accordance with the browsing history to date content of the target user is a term that varies with respect to The second term on the right side, b _pv · pv (a, t), is a term in which the value does not fluctuate according to the target user, but fluctuates according to the number of times the content is viewed at time t. Therefore, a higher score R (u, a, t) is derived as the content is similar to the content frequently browsed by the user and the change in the number of browsing times with the passage of time is small. In the mathematical formula (1), the score R (u, a, t) may be derived as the product of the first term and the second term, but the accuracy of recommending the content of interest to the target user can be determined. In order to improve, it is preferable that the score R (u, a, t) is derived as the sum of the first term and the second term.

Next, the recommendation content determination unit 116 determines the recommendation content recommended to be viewed by the target user from a set of a plurality of contents corresponding to each content vector based on the derived score (S106).

Next, the content distribution unit 118 uses the communication unit 102 to distribute one or more contents determined as the recommended content by the recommendation content determination unit 116 to the terminal device 10 of the target user (S108). This ends the processing of this flowchart.

[Processing flow during learning]
Hereinafter, a series of processing flows during learning of the information providing device 100 will be described according to a flowchart. The learning time is a state in which the learning device used at the time of operation is learned. FIG. 9 is a flowchart showing a flow of a series of processes by the information providing device 100. The processing of this flowchart may be repeated, for example, at a predetermined cycle.

_{First, the learning processing unit 122 learns the parameter b pv} included in the score derivation formula of the mathematical formula (1) (S200). For example, in the learning processing unit 122, the smaller the change in the number of views of _{the content a with the passage of time, the larger the value of the parameter bpv} , and the larger the change in the number of views of the content a with the passage of time, the larger the parameter. Learn so that the value of b _{pv becomes small.}

FIG. 10 is _{a diagram showing an example of the relationship between the parameter b pv} and the time change of the number of times the content is viewed. For example, as shown in the figure, the parameter b _pv may be represented by a step function that decreases as the time change of the number of times the content is viewed increases. For example, as the contents ID of 3 is the content of the "CONT_1" and "CONT_3" certain cycle only projecting to browse a large number of times the content case (hereinafter referred to as buzz content), the value of the parameter _{b pv} is reduced Cheap. Therefore, even if the content a for which the score is derived is buzz content and the value of the parameter pv (a, t) becomes large, the parameter b _pv reduces the value, and as a result, the first item on the right side of the mathematical formula (1). The second term becomes smaller. For example, the learning processing unit 122 determines the amount of decrease in _{the parameter bpv} according to the number of times of browsing by machine learning. The parameter b _pv is not limited to the step function, and may be any function as long as it is a function that decreases as the time change of the number of times the content is viewed increases.

Next, the learning processing unit 122 learns the parameters of the learning device defined by the learning device information 142 (S204). For example, when the user vector output by the learner is represented by the weighted average of a plurality of content vectors input to the learner, the learning processing unit 122 determines the weight given to each content vector by learning. .. Formula (2) is an example of a user vector derivation formula.

In the mathematical formula (2), w _{1 to w k} are weights given to each of the _{content vectors V cont} (1) to V _cont (k) input to the learner. These weights w _{1 to w k} are the coupling coefficients (parameters of the activation function) of the neural network when the learner is a recurrent neural network. For example, the learning processing unit 122, as illustrated in FIG. 6, if the user ID is targeted user the user's "USER_1", in deriving the user vector _{V USER} (1) of the user, content ID _{The weights w 1 to w 3 given to the content vectors V cont} (1) to V _cont (3) of the contents of "CONT_1", "CONT_2", and "CONT_3" are compared with the weights given to other content vectors. Update by error backpropagation so that it becomes larger.

At this time, the learning processing unit 122 learns so that the weight w becomes smaller as the content has a larger time change in the number of views. For example, in FIG. 3, the content the content ID "CONT_1" and "CONT_3" the content ID is compared with the content of "CONT_2", since the change in the number of views for each predetermined period _{T i} is large, Indicates that it is buzz content. Therefore, in the learning processing unit 122, since the user whose user ID is originally "USER_1" browses the contents of "CONT_1", "CONT_2", and "CONT_3", these content vectors V _cont (1). When the weight _w 1 to w ₃ for imparting the ~V _cont (3) is determined by the machine learning to be greater, because the content of the "CONT_1""CONT_3" is buzz content, content of the content of the "CONT_1" _{Both the weight w 1} given to the vector V _cont (1) and _{the weight w 3} given to the content vector V _cont (3) of the content of "CONT_3" are the content vector V of the content of "CONT_2" which is not buzz content. It is determined by machine learning so that it is smaller than the weight w _{2 given to the content (2).} As a result, when deriving the target user vector, among the content vectors of the plurality of contents viewed by the target user, the content vector (for example, V _cont (2)) of the content in which the time change of the number of views is small is the target user. The contribution to the vector can be increased, and the contribution to the target user vector can be lowered as _{the content vector (for example, V cont} (1) or V _{cont (2)) of the content whose number of views changes with time is large.} As a result, the similarity of the user vector to the buzz content becomes small, so that the buzz content can be excluded from the recommended content.

Next, the learning processing unit 122 updates the learning device information 142 stored in the storage unit 130 after updating the parameters of the learning device (S204), and ends the processing of this flowchart.

11 and 12 are diagrams schematically showing the contents of processing when determining the parameters of the learner. For example, in the example of FIG. 11, at the time of a certain period T _1, and the content A that is associated with the "election", and the content B related to "baseball" are delivered, the target user k the two content A And B are browsed. Further, at _{the time of this cycle T 1} , the time change of the number of times of viewing the content A is 0.01, and the time change of the number of times of viewing the content B is 0.9. _{In such a case, the learning processing unit 122 increases the weight of the content vector V cont} (B) of the content B in which the time change of the number of views is large among the contents viewed by the target user k, and the time change of the number of views changes. The parameters of the learner are determined so as to reduce the weight of the content vector V _{cont (A) of the small content A.}

In the example of FIG. 12, the state of content distribution at the time of the _{cycle T 2} in which the time is further advanced from the time of _{the cycle T 1 is shown.} At the time of the period T _2, we continue to have been delivered and the content B in relation to the content A related to "election", "baseball" is, furthermore, content C related to "disaster" has been newly delivered. It is assumed that the target user k has browsed the contents B and C among these contents. Further, at the time of the period T _2, the time variation of the number of views of the content A is 0.01, the time variation of the number of views the content B is 0.9, the time variation of the number of views the content C 0 It shall be 0.3. _{In such a case, the learning processing unit 122 increases the weight of the content vector V cont} (B) of the content B in which the time change of the number of views is large among the contents viewed by the target user k, and the time change of the number of views changes. The parameters of the learner are determined so as to reduce the weight of the content vector V _{cont (C) of the small content C.} As a result, when the target user k browses buzz content that is temporarily viewed frequently, such as content such as "election" and "disaster", the target user vector that is distributed and expressed by a plurality of content vectors , The contribution of buzz content is lower than that of other content. On the other hand, when the target user k browses content such as "baseball" whose number of views is substantially constant regardless of time, the contribution of the content such as "baseball" relatively increases in the target user vector. .. In this way, even if the user browses the buzz content that has been temporarily viewed a large number of times due to a temporary event such as an election being held at that time, the browsed buzz content The user vector can be generated after reducing the contribution of. As a result, it is possible to more accurately recommend the content that is expected to be of great interest to the user, as compared with the case where the contribution of the content vector of the content that is frequently viewed is simply increased.

In the above-mentioned example, the learning processing unit 122 has been described as determining the weight of each content vector according to the time change of the number of times the content is viewed when the user vector is generated, but the present invention is not limited to this. For example, when the content is posted on a medium such as a web page or an application, the learning processing unit 122 may determine the weight of the content vector according to the posting position of the content on the medium. For example, when the medium is a web page, the learning processing unit 122 reduces the weight w as the posting position of the content viewed by the user is above the web page, and the posting position of the content is below the web page. The weight w may be increased as much as possible. Thereby, for example, the user vector can be appropriately generated in consideration of the fact that the content whose posting position is higher is easier to be browsed and the content whose posting position is lower is less likely to be browsed.

According to the embodiment described above, the user vector generation unit 114 that generates a user vector in which the user's characteristics are vectorized based on the content vector in which the characteristics of the content viewed by the user are vectorized, and the user vector generation unit 114. Based on the user vector generated by the unit 114 and the content vector of each of the plurality of contents to be distributed, the recommendation content determination unit 116 that determines the recommended content to be recommended to the user from the plurality of contents to be distributed. Of the content vectors of one or more contents viewed by the user when the user vector generation unit 114 generates the user vector, the feature of the content in which the change in the number of views with the passage of time is small. By largely reflecting it in the user vector, it is possible to accurately recommend the content that is expected to be of great interest to the user.

In general, there are users who have a tendency to browse content that is often viewed in the world, content posted as an advertisement, etc., in addition to the content that they are interested in. In the case of such a user, if the contribution of the content vector of the content that is frequently viewed is simply increased to generate the user vector, the content similar to the content is recommended to the user. However, the content that is often viewed in the world and the content posted as an advertisement are not necessarily the content that the user is interested in, and the content that is not of interest to the user may be recommended.

On the other hand, in the present embodiment, the number of views is adjusted in order to adjust the contribution of the content vector according to the magnitude of the change in the number of views over time, regardless of whether the number of views is high or low. If the content with a large number of views is not uniformly judged to be of interest to the user, and the change in the number of views over time is small even for the content with a small number of views, the user is essentially the content. It can be judged that the content is of interest. This makes it possible to generate an appropriate user vector. As a result, even if a user accidentally browses content that is momentarily viewed by an unspecified number of users, such as a gossip article, the gossip article is given to that user. It is possible to make it difficult to recommend such content, and it is possible to recommend content that the user is expected to be really interested in.

<Hardware configuration>
The information providing device 100 of the above-described embodiment is realized by, for example, a hardware configuration as shown in FIG. FIG. 13 is a diagram showing an example of the hardware configuration of the information providing device 100 of the embodiment.

The information providing device 100 includes NIC100-1, CPU100-2, RAM100-3, ROM100-4, a secondary storage device 100-5 such as a flash memory or an HDD, and a drive device 100-6 having an internal bus or a dedicated communication line. The configuration is interconnected by. A portable storage medium such as an optical disk is mounted on the drive device 100-6. A program stored in a portable storage medium mounted on the secondary storage device 100-5 or the drive device 100-6 is expanded into the RAM 100-3 by a DMA controller (not shown) or the like, and executed by the CPU 100-2. As a result, the control unit 110 is realized. The program referred to by the control unit 110 may be downloaded from another device via the network NW.

Although the embodiments for carrying out the present invention have been described above using the embodiments, the present invention is not limited to these embodiments, and various modifications and substitutions are made without departing from the gist of the present invention. Can be added.

1 ... Information providing system, 10 ... Terminal device, 100 ... Information providing device, 102 ... Communication unit, 110 ... Control unit, 112 ... Content vector generation unit, 114 ... User vector generation unit, 116 ... Recommended content determination unit, 118 ... Content distribution unit, 120 ... information update unit, 122 ... learning processing unit, 130 ... storage unit

Claims (9)

A derivation unit that derives the characteristics of the user based on the characteristics of the content viewed by the user, and
Based on the characteristics of the user derived by the derivation unit and the characteristics of each of the plurality of contents to be distributed, a determination unit that determines the content recommended to the user from the plurality of contents to be distributed. , Equipped with
The derivation unit derives the characteristics of the user based on the change in the number of views of each content viewed by the user over time.
Information providing device. When the derivation unit derives the characteristics of the user, among the characteristics of one or more contents viewed by the user, the characteristics of the content in which the change in the number of views with the passage of time is small, the larger the characteristics of the user. Reflect in
The information providing device according to claim 1. The derivation unit is configured to give weights to the characteristics of one or more input contents and output the characteristics of the user who has viewed the contents based on the characteristics of the one or more contents to which the weights are given. By inputting the characteristics of the content into the learning device, the characteristics of the user are derived.
The information providing device according to claim 1 or 2. A learning processing unit for learning the learner is further provided so that the feature of the content in which the change in the number of views of the content is small with the passage of time is given a larger weight.
The information providing device according to claim 3. The decision unit
Derivation of the similarity between the characteristics of the user derived by the derivation unit and the characteristics of each of the plurality of contents to be distributed is derived.
Based on the similarity derived for each feature of the plurality of contents to be distributed and the index value based on the number of views of each of the plurality of contents to be distributed, the plurality of contents to be distributed Determine the content to recommend to the user from among
The information providing device according to any one of claims 1 to 4. A learning processing unit that determines the index value by machine learning according to a change in the number of views of the content with the passage of time is further provided.
The information providing device according to claim 5. The learning processing unit determines that the index value is reduced as the change in the number of views of the content with the passage of time is larger, and is increased as the change in the number of views of the content with the passage of time is smaller.
The information providing device according to claim 6. The computer
Based on the characteristics of the content viewed by the user, the characteristics of the user are derived, and the characteristics of the user are derived.
Based on the characteristics of the derived user and the characteristics of each of the plurality of contents to be distributed, the content recommended to the user is determined from the plurality of contents to be distributed.
Based on the change in the number of views of each content viewed by the user over time, the characteristics of the user are derived.
Information provision method. On the computer
The process of deriving the characteristics of the user based on the characteristics of the content viewed by the user, and
A process of determining the content recommended to the user from the plurality of contents to be distributed based on the derived characteristics of the user and the characteristics of each of the plurality of contents to be distributed.
A process of deriving the characteristics of the user based on a change in the number of views of each content viewed by the user over time.
A program to execute.

Images