JP6986943B2 - Content evaluation device, content distribution system, content evaluation method, and program - Google Patents

Description

The present invention relates to a content evaluation device, a content distribution system, a content evaluation method, and a program.

Content such as news articles provided to users via the network is evaluated based on the content, and based on the evaluation result, whether or not the content is posted on the Internet site etc. and the order of posting are determined. There is. For example, in the invention described in Patent Document 1, the publication order is determined based on the importance and information freshness of the article information.

Japanese Unexamined Patent Publication No. 2017-59057 Japanese Unexamined Patent Publication No. 2016-29536 JP-A-2015-19771 Japanese Unexamined Patent Publication No. 2009-3740

The inventions described in Patent Documents 2 to 4 describe measuring or predicting the time for a user to view a content, but the information is used in connection with the presence or absence of the content and the order of publication. Is not mentioned. For this reason, with the conventional technology, it may not be possible to evaluate the content according to the evaluation criteria according to the needs of the distributor.

The present invention has been made in consideration of such circumstances, and is a content evaluation device, a content distribution system, a content evaluation method, and a program capable of evaluating content according to evaluation criteria according to the needs of a distributor. It is one of the purposes to provide.

One aspect of the present invention is an acquisition unit that acquires a first evaluation result that evaluates content distributed to a user via a network based on a predetermined rule, and the first evaluation result acquired by the acquisition unit. The content evaluation device includes an evaluation unit for deriving a second evaluation result based on a predicted stay time, which is a time expected to be viewed by a user, as an index when distributing the content.

According to one aspect of the present invention, it is possible to provide a content evaluation device, a content distribution system, a content evaluation method, and a program capable of evaluating content according to evaluation criteria according to the needs of a distributor.

It is a figure which shows an example of the use environment and the structure of the content distribution system 100 using the content evaluation apparatus (content evaluation unit 110) of 1st Embodiment. It is a figure which shows typically the content of the processing by the 1st evaluation unit 112. It is a figure which shows an example of the relationship between the number of characters of a content, and the predicted stay time. It is a flowchart which shows an example of the flow of the process executed by the content evaluation unit 110 of 1st Embodiment. It is a flowchart which shows an example of the flow of the process executed by the content selection unit 140 and the content distribution unit 150 of the first embodiment. It is a figure which shows an example of the use environment and the structure of the content distribution system 200 which used the content evaluation apparatus (content evaluation unit 220) of 2nd Embodiment. It is a figure which shows an example of the structure and function of the 1st evaluation part 222 of 2nd Embodiment. It is a flowchart which shows an example of the flow of the partial processing executed by the content evaluation unit 220 of the 2nd Embodiment. It is a flowchart which shows an example of the flow of the process executed by the content distribution system 200 of 2nd Embodiment. It is a figure which shows an example of the use environment and the structure of the content distribution system 300 which used the content evaluation apparatus (mass content evaluation unit 310) of 3rd Embodiment.

<Overview>
Hereinafter, embodiments of the content evaluation device, content distribution system, content evaluation method, and program of the present invention will be described with reference to the drawings. The content evaluation device is a device that evaluates content distributed to users via a network. The content evaluation device and the content distribution system are realized by one or more processors. The content is, for example, a news article, but may be other content such as a video. In the following description, the content is assumed to be a news article. In addition, if it conveys any information regardless of whether it is news or not, it is considered to be a news article. News articles contain at least textual information, so the number of characters can be counted. In addition, news articles may include non-text items such as still images and moving images, figures (switches) that function as interfaces, decorative effects, and the like.

The content evaluation device derives a first evaluation result in which the content is evaluated based on a predetermined rule, and a second evaluation result based on the predicted stay time, which is the time when the user is expected to view the content. This second evaluation result is used as an index for determining whether or not the content is posted and the order in which the content is posted. The content evaluation device may acquire the first evaluation result by deriving it by itself, or may acquire the first evaluation result from another device. Further, the content evaluation device may be integrated with the device for distributing the content, or may be configured as a separate body. In the following embodiments, the content evaluation device shall function as a content distribution system by being integrated with the device for distributing the content.

<First Embodiment>
[composition]
Hereinafter, the first embodiment will be described. FIG. 1 is a diagram showing an example of a usage environment and configuration of a content distribution system 100 using the content evaluation device (content evaluation unit 110) of the first embodiment.

The content distribution system 100 communicates with the user's terminal device 10 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.

The terminal device 10 is a terminal device including an input device, a display device, a communication device, a storage device, and an arithmetic unit, 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 browser or an application program is activated, and a request according to the content input by the user is transmitted to the content distribution system 100. Further, the UA displays various images based on the information acquired from the content distribution system 100.

The content distribution system 100 functions as a web server that provides a web page to the terminal device 10 in response to a request from a browser, or an application server that provides an image or sound in response to a request from an application program. do. The content distribution system 100 includes a network card such as a NIC.

The content distribution system 100 includes, for example, a content evaluation unit 110, a content selection unit 140, a content distribution unit 150, and a content performance analysis unit 160. The content evaluation unit 110 includes a first evaluation unit 112 and a second evaluation unit 114. These components are realized by, for example, a hardware processor such as a CPU (Central Processing Unit) executing a program (software). In addition, some or all of these components are hardware (circuits) such as LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), and GPU (Graphics Processing Unit). It may be realized by the part; including circuitry), or it may be realized by the cooperation of software and hardware.

The content distribution system 100 refers to the information stored in the storage unit 190, and performs processing while appropriately writing the processing result to the storage unit 190. The storage unit 190 is realized by a RAM (Random Access Memory), a ROM (Read Only Memory), an HDD (Hard Disk Drive), a flash memory, or the like. The storage unit 190 may store a program executed by the hardware processor of the content distribution system 100. The storage unit 190 may be attached to the content distribution system 100 (included in the content distribution system 100 or the content evaluation unit 110), or may be NAS (Network Attached) accessible to the content distribution system 100 via the network NW. It may be an external device such as Storage). In addition to the above program, the storage unit 190 stores, for example, information such as content information 191 and user information 192, action log 193, content performance information 194, and learning model information 195.

The content information 191 includes text information, images, audio and the like included in the news article, and information such as the distribution source and the author of the news article. The user information 192 includes a user ID which is identification information of the user, demographic information of the user (meaning age, gender, occupation, address and other information) and the like. The action log 193 is information based on the browsing history of the content by the user. The content record information 194 is information based on the browsing record for each content. The learning model information 195 is information that defines a learning model learned by the first evaluation unit 112. Details of these will be added below as appropriate.

[First Evaluation Department]
The first evaluation unit 112 derives, for example, the estimated importance (an example of the first evaluation result) of evaluating the content based on a predetermined rule that does not depend on the attributes of the user. FIG. 2 is a diagram schematically showing the contents of processing by the first evaluation unit 112. For example, the first evaluation unit 112 obtains the estimated importance using only the content information 191 for the first few times, and when the past actual value starts to be input, the RNN (Recurrent) makes a correction using it. Neural Network). In addition to the rules described below, the predetermined rule may be a rule in which some element depending on the user's attribute is added.

First, the first evaluation unit 112 obtains the _{distributed expression y 0} from the sparse vector x of the content information related to the news article at the stage when the news article is submitted and stored in the storage unit 190 as the content information 191. Such processing is represented by the formula (1). In the equation, A is a matrix and a is a vector. The function f () is represented by the equation (2). That is, the function f () is a function that replaces each element of the vector with zero if it is negative.

As described above, the content information 191 includes text information, images, audio and the like included in the news article, and information such as the distribution source and the author of the news article. The first evaluation unit 112 vectorizes text information by using techniques such as tf-idf (Term Frequency-Inverse Document Frequency) and word2vec. In addition, for images and sounds, feature quantities in vector format are obtained based on various techniques. The first evaluation unit 112 integrates (for example, adds) the vector data obtained from the above-mentioned various information, and obtains the sparse vector x of the content information.

The first evaluation unit 112 repeatedly executes the following processing with the passage of time.

The first evaluation unit 112, a k-th process in which is repeatedly executed, wherein the information y _k-1 obtained in (k-1) -th processing, the times of the time information t _k and actual information h _k Based on the above, the _{process of obtaining the feature information y k} is performed. Such processing is represented by the formula (3). The details of the actual information will be described later. In the equation, B is a matrix, b is a vector, and T is a transpose. The first evaluation unit 112 _{multiplies the matrix obtained by arranging the transposed vector of the feature information y k-1} , the transposed vector of the time information t _k , and the transposed vector of the actual information h _k , and the matrix B, and adds the vector b. Therefore, the feature information y _k is obtained.

Next, the first evaluation unit 112 calculates the _{estimated feature z k} at the time “k” based on the _{feature information y k.} Such processing is represented by the formula (4). In the equation, C is a matrix and c is a vector. The function σ () is a sigmoid function represented by the equation (5).

Next, the first evaluation unit 112 calculates the estimated importance w _{k based on the estimated feature z k.} Such processing is represented by the formula (6). In the equation, D is a matrix and d is a vector.

In this way, the repetition of the latest of the estimated severity w _k obtained what is treated as an estimate importance wk of the content.

The first evaluation unit 112 learns the values of the matrices A to D and the values of the vectors a to d by using a method such as a stochastic gradient descent (SGD) method. Then, the first evaluation unit 112 corrects the connection coefficient in the RNN based on the learning result. Thereby, it is possible to improve the estimation accuracy of the estimated severity w _k. The learned connection coefficient is stored in the storage unit 190 as a part of the learning model information 195. The learning model information 195 is information including connection information of neurons, connection coefficients, and the like.

[About achievement information]
Hereinafter, the actual information h _k will be described. The content distribution unit 150 accumulates the action log 193 when distributing the content. It should be noted that such processing may be performed by a log collecting device separate from the content distribution system 100. The action log 193 is, for example, information indicating a browsing history of contents for each user. The content performance analysis unit 160 generates the content performance information 194 based on the action log 193. The content performance information 194 is, for example, information such as the number of impressions, the number of clicks, and the number of tweets for each content and for each time zone when looking back up to a predetermined time in the past. Content record information 194 corresponding to a certain content is handled as record information h _k described above. For example, the actual information h _k is a vectorization of the number of impressions for each time zone when looking back up to the past predetermined time of the target content.

In performing the first evaluation unit 112 repeats the processing, immediately after trafficking news article is content, record information h _k is zero vector. Therefore, immediately after the submission, the estimated importance _wk is derived exclusively based on the content information 191. After that, when the actual information h _k begins to have a significant value, the influence of the actual information h _{k on the estimated importance w k} becomes large. As a result, the estimated importance _wk (first evaluation result) can be derived more flexibly.

[Second evaluation department]
The second evaluation unit 114, the content of interest, estimation and severity w _k derived by the first evaluation unit 112 derives the second evaluation results based on the predicted residence time. The estimated stay time is the time when the user is expected to view the target content. The second evaluation unit estimates the predicted staying time so that the predicted staying time becomes longer as the number of characters in the content increases, for example. However, if the number of characters in the content becomes too large, the user may stop viewing in the middle, so the second evaluation unit 114 predicts that the rate of change in the predicted stay time will decrease as the number of characters in the content increases. You may estimate the length of stay. FIG. 3 is a diagram showing an example of the relationship between the number of characters in the content and the estimated staying time.

In addition, the presence or absence of an image, the characteristics of the image, and the like may be taken into consideration as other factors in estimating the predicted stay time. For example, the second evaluation unit 114 indicates that the image or the content is provided with information indicating that the image is an image that simply suggests the content of the news article as a characteristic of the image, as compared with the case where the image or the content is not provided. Estimate the estimated staying time short. On the contrary, when the image or the content is provided with information indicating that the image has little relation to the content of the news article as a characteristic of the image, the predicted stay time is estimated to be longer than when the image or the content is not provided. You may. In either case, the second evaluation unit 114 may increase the correction amount for shortening or lengthening the predicted stay time as the display position of the image is on the upper side with respect to the display surface of the content.

The second evaluation unit 114 integrates the above trends to estimate the predicted dwell time. The method for estimating the predicted dwell time ct by the second evaluation unit 114 is expressed by, for example, the equation (7). g () is an arbitrary function for realizing the above tendency.

tk = g {(number of characters), (presence / absence of image), (characteristics of image), (position of image)} ... (7)

The second evaluation unit 114, for example, by multiplying the predicted residence time tc to the estimated importance w _k, derives the second evaluation result SC to be used as an indicator of when delivering content. Such processing is represented by the formula (8). The second evaluation result SC is high, it is estimated that the click rate of the user is high because the importance of the content itself is high, and when the content is started to be viewed, the viewing is continued for a relatively long time. Means that. Therefore, by preferentially delivering content with a long second evaluation result SC, it is possible to improve user satisfaction, promote retention in the service, and acquire a large amount of user location information. You can expect the effect of doing. These effects may depend on the needs of the content distributor. Therefore, according to the content evaluation unit 110, the content can be evaluated according to the evaluation criteria according to the needs of the distributor.

SC = w _k × tk… (8)

[Content distribution]
The content selection unit 140 selects the content to be delivered to the user based on the second evaluation result SC. For example, the content selection unit 140 selects the contents in descending order of the second evaluation result SC, and mixes the randomly selected contents at a fixed ratio. The content distribution unit 150 distributes the content selected by the content selection unit 140 to the user's terminal device 10.

[Processing flow]
FIG. 4 is a flowchart showing an example of the flow of processing executed by the content evaluation unit 110 of the first embodiment. The processing of this flowchart is repeatedly executed. First, the content evaluation unit 110 determines whether or not the evaluation timing has arrived (S100). The evaluation timing may be, for example, a timing at which the news article arrives on a regular basis, or may include a timing at which a new news article is newly submitted.

When the evaluation timing arrives, the content evaluation unit 110 performs the processes of S110 to S118 for each target content. The target content may be all content, or may be limited to the submitted content at the timing when a new news article is submitted. In addition, old contents may be excluded from the target as appropriate.

First, the content performance analysis unit 160 updates the content performance information 194 (S110). The process of S110 may be executed asynchronously with the process of this flowchart.

Next, first evaluating unit 112 derives a first evaluation value (estimated importance _{w k)} (S112). Next, the second evaluation unit 114 derives the predicted residence time ct of the target content or reads it from the storage unit 190 (S114). When the predicted stay time ct does not change with time, the second evaluation unit 114 stores the result in the storage unit 190 once the predicted stay time ct is derived for a certain content, and in the subsequent processing. It may be read from the storage unit 190.

Next, the second evaluation unit 114 derives the second evaluation result SC (S116). The content evaluation unit 110 stores various information obtained in the process of the above process and the second evaluation result SC in the storage unit 190 (S118). This ends the processing of one routine in this flowchart.

FIG. 5 is a flowchart showing an example of the flow of processing executed by the content selection unit 140 and the content distribution unit 150 of the first embodiment. First, the content distribution unit 150 determines whether or not there is a content request (S120).

When there is a request for content, the content selection unit 140 selects the content based on the second evaluation result SC as described above (S130), and the content distribution unit 150 selects the content selected in S130 as the user's terminal. Deliver to device 10 (S132). Then, the content distribution unit 150 updates the action log of the user (S134). This ends the processing of one routine in this flowchart.

According to the content evaluation unit 110 (content evaluation device) of the first embodiment described above, the first evaluation result of evaluating the content distributed to the user via the network NW is derived based on a predetermined rule ( The second evaluation unit 114 that derives the first evaluation unit 112 (acquired), the first evaluation result derived by the first evaluation unit 112, and the second evaluation result based on the predicted dwell time as an index when distributing the content. By providing, the content can be evaluated according to the evaluation criteria according to the needs of the distributor.

The method for deriving the first evaluation result in the first embodiment is merely an example. The first evaluation result may be derived by other rules.

<Second Embodiment>
Hereinafter, the second embodiment will be described. FIG. 6 is a diagram showing an example of a usage environment and configuration of a content distribution system 200 using the content evaluation device (content evaluation unit 220) of the second embodiment.

The content distribution system 200 communicates with the user's terminal device 10 via the network NW. Since the terminal device 10 is the same as that of the first embodiment, the description thereof will be omitted. The content distribution system 200 functions as a web server that provides a web page to the terminal device 10 in response to a request from a browser, or an application server that provides an image or sound in response to a request from an application program. do. The content distribution system 200 includes a network card such as a NIC.

The content distribution system 200 includes, for example, a content evaluation unit 220, a content selection unit 240, a content distribution unit 250, and a content performance analysis unit 260. The content evaluation unit 220 includes a first evaluation unit 222 and a second evaluation unit 224. These components are realized by, for example, a hardware processor such as a CPU executing a program (software). Further, some or all of these components may be realized by hardware such as LSI, ASIC, FPGA, GPU (circuit unit; including circuitry), or realized by collaboration between software and hardware. May be done.

The content distribution system 200 refers to the information stored in the storage unit 290, and performs processing while appropriately writing the processing result to the storage unit 290. The storage unit 290 is realized by a RAM, a ROM, an HDD, a flash memory, or the like. The storage unit 290 may store a program executed by the hardware processor of the content distribution system 200. The storage unit 290 may be attached to the content distribution system 200 (included in the content distribution system 200 or the content evaluation unit 220), or may be external to the content distribution system 200 such as NAS accessible via the network NW. It may be a device. In addition to the above program, the storage unit 290 stores, for example, information such as content information 291, user information 292, action log 293, content performance information 294, and learning model information 295.

The content information 291 includes text information, images, audio and the like included in the news article, and information such as the distribution source and the author of the news article. The user information 292 includes a user ID, which is user identification information, user demogra information, and the like. The action log 293 is information based on the browsing history of the content by the user. The content record information 294 is information based on the browsing record for each content. The learning model information 295 is information that defines a learning model learned by the first evaluation unit 222. Details of these will be added below as appropriate.

[First Evaluation Department]
FIG. 7 is a diagram showing an example of the configuration and function of the first evaluation unit 222 of the second embodiment. The first evaluation unit 222 includes a content vector generation unit 222A, a user vector generation unit 222B, and a similarity determination unit 222C. The first evaluation unit 222 evaluates the content according to a rule depending on the attribute based on the user's action log as described below.

The content vector generation unit 222A first performs morphological analysis or the like on the content of the content, and divides the text information into words (morphemes). Next, the content vector generation unit 222A applies a method such as word2vec to each word to acquire a distributed expression word vector. One or both of the morphological analysis and word2vec may be realized by requesting the processing to an external device instead of the content vector generation unit 222A performing the processing by itself.

Then, the content vector generation unit 222A derives the content vector Vc by inputting one or more word vectors acquired from the content into the content vector derivation model. The content vector derivation model is, for example, a model generated by machine learning. The content vector derivation model is generated by finding the parameter (connection coefficient) of the activation function of the hidden layer so that the relationship between the known word vector and the content vector Vc can be reproduced by a method such as deep learning. To.

Notwithstanding the above definition, the content vector derivation model may include the identification of word2vec, that is, what kind of corpus is given to word2vec.

The content vector generation unit 222A stores the generated content vector Vc together with the content ID in the storage unit 290 as a part of the learning model information 295.

The user vector generation unit 222B refers to the action log 293, acquires the content ID of the content viewed by the user, and acquires the content vector Vc corresponding to the acquired content ID from the storage unit 290. The user vector generation unit 222B derives the user vector Vu by inputting the acquired one or more content vectors Vc into the user vector derivation model. The user vector derivation model is, for example, a model generated by machine learning. In the user vector derivation model, for example, the parameter (connection coefficient) of the activation function of the hidden layer is obtained so that the relationship between the known content vector Vc and the user vector Vu can be reproduced by a method such as deep learning. Generated by. The user vector generation unit 222B stores the generated user vector Vu together with the user ID in the storage unit 290 as a part of the learning model information 295. The learning model information 295 further includes information defining a content vector derivation model and a user vector derivation model (for example, their neuron connection information and connection coefficients). The user vector Vu may be created by reflecting elements other than the content vector Vc of the browsed content. The elements other than the content vector Vc are, for example, a part or all of a query input by the user, a browsing history of a web page that does not correspond to the content, demogra information of the user, and the like.

The similarity determination unit 222C derives, for example, the cosine similarity Sm between the user vector Vu and the content vector Vc as the first evaluation result of evaluating the content for each user for the target user.

[Second evaluation department]
The second evaluation unit 224 is used as an index when the content is distributed by multiplying the cosine similarity Sm, which is the first evaluation result, by the predicted stay time ct, similarly to the second evaluation unit 114 of the first embodiment. The second evaluation result SC * is derived. The second evaluation result SC * is the result of evaluating the content according to the rule depending on the user's attribute (behavior log).

[Content distribution]
The content selection unit 240 selects the content to be delivered to the user based on the second evaluation result SC *. For example, the content selection unit 240 selects the contents in descending order of the second evaluation result SC *, and mixes the randomly selected contents at a fixed ratio. The content distribution unit 250 distributes the content selected by the content selection unit 240 to the user's terminal device 10.

[Processing flow]
FIG. 8 is a flowchart showing an example of a partial processing flow executed by the content evaluation unit 220 of the second embodiment. First, the content evaluation unit 220 determines whether or not the content has been submitted (S200).

When the content is submitted, the content vector generation unit 222A of the first evaluation unit 222 derives the content vector Vc and stores it in the storage unit 290 (S210). Further, the second evaluation unit 224 derives the predicted stay time of the content and stores it in the storage unit (S212). This ends one routine in this flowchart.

FIG. 9 is a flowchart showing an example of a processing flow executed by the content distribution system 200 of the second embodiment.

First, the content distribution unit 250 determines whether or not there is a content request (S220).

When there is a request for content, the content selection unit 240 selects the content based on the second evaluation result SC * (S230), and the content distribution unit 250 transfers the content selected in S230 to the user's terminal device 10. Deliver (S232). Then, the content distribution unit 250 updates the action log of the user (S234).

Next, the user vector generation unit 222B generates the user vector Vu based on the updated action log (S236). Next, the similarity determination unit 222C makes a first evaluation result (cosine similarity Sm) for each content based on the user vector Vu generated in S236 and the content vector Vc of each content already generated. Is derived (S238).

Next, the second evaluation unit 224 derives the second evaluation result SC * for each content based on the first evaluation result derived in S238 and the predicted stay time derived in advance (S240), and the second evaluation result SC * is derived. 2 The evaluation result SC * is stored in the storage unit 290 (S242). This ends the processing of one routine in this flowchart.

According to the content evaluation unit 210 (content evaluation device) of the second embodiment described above, the same effect as that of the first embodiment can be obtained.

<Third Embodiment>
Hereinafter, the third embodiment will be described. FIG. 10 is a diagram showing an example of a usage environment and configuration of a content distribution system 300 using the content evaluation device (mass content evaluation unit 310) of the third embodiment.

The content distribution system 300 communicates with the user's terminal device 10 via the network NW. Since the terminal device 10 is the same as that of the first embodiment, the description thereof will be omitted. The content distribution system 300 functions as a web server that provides a web page to the terminal device 10 in response to a request from a browser, or an application server that provides an image or sound in response to a request from an application program. do. The content distribution system 300 includes a network card such as a NIC.

The content distribution system 300 includes, for example, a mass content evaluation unit 310, a content evaluation unit 322 for each user, a content selection unit 340, a content distribution unit 350, and a content performance analysis unit 360. The mass content evaluation unit 310 includes a first evaluation unit 312 and a second evaluation unit 314. These components are realized by, for example, a hardware processor such as a CPU executing a program (software). Further, some or all of these components may be realized by hardware such as LSI, ASIC, FPGA, GPU (circuit unit; including circuitry), or realized by collaboration between software and hardware. May be done.

The content distribution system 300 refers to the information stored in the storage unit 390, and performs processing while appropriately writing the processing result to the storage unit 290. The storage unit 390 is realized by a RAM, a ROM, an HDD, a flash memory, or the like. The storage unit 390 may store a program executed by the hardware processor of the content distribution system 300. The storage unit 390 may be attached to the content distribution system 300 (provided by the content distribution system 300, the mass content evaluation unit 310, or the content evaluation unit 322 for each user), or the content distribution system 300 is a network. It may be an external device such as NAS that can be accessed via the NW. In addition to the above program, the storage unit 390 stores information such as content information 391, user information 392, action log 393, content performance information 394, and learning model information 395.

The content information 391 includes text information, images, audio and the like included in the news article, and information such as the distribution source and the author of the news article. The user information 392 includes a user ID, which is user identification information, user demogra information, and the like. The action log 393 is information based on the browsing history of the content by the user. The content record information 394 is information based on the browsing record for each content. The learning model information 395 is information that defines a learning model that is learned by the first evaluation unit 312 and the content evaluation unit 322 for each user. Details of these will be added below as appropriate.

The first evaluation unit 312 and the second evaluation unit 314 of the mass content evaluation unit 310 have the same functions as the first evaluation unit 112 and the second evaluation unit 114 of the content evaluation unit 110 according to the first embodiment. That is, mass content evaluation unit 310 multiplies the estimated importance w _k to the prediction residence time tc, derives a second evaluation result SC to be used as an indicator of when delivering content.

The content evaluation unit 322 for each user has the same function as the first evaluation unit 222 of the content evaluation unit 220 according to the second embodiment. That is, the content evaluation unit 322 for each user derives the cosine similarity Sm between the user vector Vu and the content vector Vc as an index for evaluating the content for each user for the target user.

The content selection unit 340 according to the third embodiment selects a content group in which the content having the higher value of the second evaluation result SC and the content having the higher value of the cosine similarity Sm are mixed at a predetermined ratio. do. The content distribution unit 350 distributes the content group selected by the content selection unit 340 to the user's terminal device 10. The function of the content performance analysis unit 360 is the same as the function of the content performance analysis unit 160 according to the first embodiment.

According to the mass content evaluation unit 310 (content evaluation device) of the third embodiment described above, the same effect as that of the first embodiment can be obtained.

The content selection unit 340 selects a content group in which the content having the higher value of the second evaluation result SC and the content having the higher value of the cosine similarity Sm are mixed at a predetermined ratio. However, not limited to this, depending on the page, site, service, domain, etc., either the content having the higher value of the second evaluation result SC or the content having the higher value of the cosine similarity Sm can be selected. You may choose. Further, the content selection unit 340 selects the content having the higher value of the second evaluation result SC at the stage where the user's action log is not sufficiently accumulated, and responds to the accumulation of the user's action log. The ratio of the content having the higher cosine similarity Sm value may be increased. Further, the content selection unit 340 may select the content by adding a random element.

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.

100, 200, 300 Content distribution system 110, 220 Content evaluation unit 112, 222, 312 First evaluation unit 114, 224, 314 Second evaluation unit 140, 240, 340 Content selection unit 150, 250, 350 Content distribution unit 160, 260, 360 Content performance analysis unit 190, 290, 390 Storage unit 191, 291, 391 Content information 192, 292, 392 User information 193, 293, 393 Action log 194, 294, 394 Content performance information 195, 295, 395 Learning Model Information 310 Mass Content Evaluation Department 322 Content Evaluation Department for each user

Claims (7)

An acquisition unit that acquires the first evaluation result that evaluates the content delivered to the user via the network based on a predetermined rule.
An evaluation that derives the first evaluation result acquired by the acquisition unit and the second evaluation result based on the predicted stay time, which is the time expected for the user to view the content, as an index when distributing the content. Department and
Equipped with
The evaluation unit derives the predicted dwell time based on the position of the image included in the content.
Content evaluation device. The evaluation unit derives the predicted dwell time based on the number of characters contained in the content.
The content evaluation device according to claim 1. The predetermined rule is a rule that does not depend on the user's attributes.
The content evaluation device according to claim 1 or 2. It also has a content evaluation unit for each user who evaluates content according to rules that depend on user attributes.
The second evaluation result and the evaluation result of the content evaluation unit for each user are output as an index for distributing the content.
The content evaluation device according to claim 3. The content evaluation device according to any one of claims 1 to 4, and the content evaluation device.
A selection unit that selects content based on the evaluation result of the content evaluation device, and
A distribution unit that distributes the content selected by the selection unit to the user's terminal device, and a distribution unit.
Content distribution system with. The computer
Obtain the first evaluation result that evaluated the content delivered to the user via the network based on the predetermined rule.
The acquired first evaluation result and the second evaluation result based on the predicted stay time, which is the time when the user is expected to view the content, are derived as an index when the content is distributed .
In deriving the index, the predicted dwell time is derived based on the position of the image included in the content.
Content evaluation method. On the computer
Based on a predetermined rule, the first evaluation result of evaluating the content delivered to the user via the network is acquired.
The acquired first evaluation result and the second evaluation result based on the predicted stay time, which is the time when the user is expected to view the content, are derived as an index when the content is distributed .
In deriving the index, the predicted dwell time is derived based on the position of the image included in the content.
program.

Images