Message playback optimization method, device, equipment and readable storage medium

Abstract

The present application discloses a message playback optimization method, device, equipment and readable storage medium, the message playback optimization method comprising: obtaining group monitoring information, and inputting the group monitoring information into a preset feature extraction model, establishing a group portrait corresponding to the group monitoring information, obtaining message context features, and inputting the message context features and the group portrait into a preset message carousel prediction model, matching the message context features and the message carousel information corresponding to the group portrait, and playing the message according to the message carousel information. The present application solves the technical problem of low efficiency and poor effect of message carousel.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology in financial technology (Fintech), and in particular to a message playback optimization method, device, equipment and readable storage medium. Background Art

[0002] With the continuous development of financial technology, especially Internet technology finance, more and more technologies (such as distributed, blockchain, artificial intelligence, etc.) are applied in the financial field, but the financial industry also puts forward higher requirements on technology, such as higher requirements on the distribution of corresponding to-do items in the financial industry.

[0003] With the continuous development of computer software and artificial intelligence, the application of machine learning models is becoming more and more extensive. At present, offline messages with multiple people coexisting are very common, such as messages in building elevators, messages on shopping mall guide screens, etc. These offline messages are usually just simply rotated on the message screen to push messages to users, but in multiple coexisting scenarios, there are usually one or more user groups, such as family groups, couples groups, etc., and in offline message scenarios, the mobility of people is very large, and user groups often change. Therefore, the messages rotated on the message screen often do not match the preferences of the user group, which leads to poor message carousel effect, and for messages that do not match the user group, the user group usually does not pay attention, which leads to low efficiency of message carousel. Therefore, there are technical problems in the prior art that offline message carousel is low in efficiency and poor in effect. Summary of the invention

[0004] The main purpose of the present application is to provide a message playback optimization method, device, equipment and readable storage medium, aiming to solve the technical problems of low efficiency and poor effect of offline message carousel in the prior art.

[0005] To achieve the above-mentioned purpose, the present application provides a message playback optimization method, which is applied to a message playback optimization device, and the message playback optimization method includes:

[0006] Acquire group monitoring information, and input the group monitoring information into a preset feature extraction model to establish a group portrait corresponding to the group monitoring information;

[0007] Acquire message context features, and input the message context features and the group portrait into a preset message carousel prediction model, and match the message context features and the message carousel information corresponding to the group portrait;

[0008] Messages are played according to the message carousel information.

[0009] Optionally, the step of inputting the message context feature and the group portrait into a preset message carousel prediction model, and matching the message carousel information corresponding to the message context feature and the group portrait comprises:

[0010] Obtaining the message context features and the model input feature representation matrix corresponding to the group portrait;

[0011] The model input feature representation matrix is ​​input into the preset message carousel prediction model, the message carousel sequence that meets the preset target message revenue is calculated, and the message carousel information is obtained.

[0012] Optionally, the step of inputting the model input feature representation matrix into the preset message carousel prediction model, calculating the message carousel sequence that meets the preset target message revenue, and obtaining the message carousel information includes:

[0013] Inputting the model input feature representation matrix into the preset message carousel prediction model to match the set of messages to be pushed corresponding to the model input feature representation matrix;

[0014] Obtain one or more message carousel sequences corresponding to the message set to be pushed, and calculate the message revenue corresponding to each message carousel sequence;

[0015] A target message benefit that meets the preset target message benefit is selected from each of the message benefits, and the target message carousel sequence corresponding to the target message benefit is used as the message carousel information.

[0016] Optionally, the preset feature extraction model includes a user individual feature extraction model and a user group feature extraction model.

[0017] The step of acquiring the group monitoring information, inputting the group monitoring information into a preset feature extraction model, and establishing a group portrait corresponding to the group monitoring information includes:

[0018] Inputting the group monitoring information into the user individual feature extraction model, extracting the user individual features in the group monitoring information, and obtaining the user individual feature recognition result;

[0019] The user individual feature recognition result is input into the user group feature extraction model, the user group feature corresponding to the user individual feature recognition result is extracted, and the group portrait is obtained.

[0020] Optionally, the user individual feature extraction model includes a user coarse-grained feature recognition model and a user interaction mode recognition model, the user individual feature recognition result includes a user coarse-grained feature recognition result and a user interaction mode recognition result, and the user individual features include user coarse-grained features and user interaction modes of one or more target users.

[0021] The step of inputting the group monitoring information into the user individual feature extraction model, extracting the user individual features in the group monitoring information, and obtaining the user individual feature recognition result comprises:

[0022] Inputting the group monitoring information into the user coarse-grained feature recognition model, extracting the user coarse-grained features of each of the target users, and obtaining the user coarse-grained feature recognition results;

[0023] The group monitoring information is input into the user interaction mode recognition model to recognize the user interaction mode between the target users, and obtain the user interaction mode recognition result.

[0024] Optionally, the user individual feature recognition result corresponds to one or more target users, and the user group feature includes a user group division result and a group social relationship recognition result.

[0025] The step of inputting the user individual feature recognition result into the user group feature extraction model, extracting the user group feature corresponding to the user individual feature recognition result, and obtaining the group portrait comprises:

[0026] Inputting the user individual feature recognition result into the user group feature extraction model, grouping each of the target users, and obtaining the user group division result, wherein the user group division result includes one or more user groups;

[0027] Identify the social relationship of each user group and obtain the group social relationship identification result;

[0028] The group portrait is established based on the user group division results and the group social relationship identification results.

[0029] Optionally, after the step of playing the message according to the message carousel information, the step includes:

[0030] Collecting message playing information and message push results corresponding to the message playing information;

[0031] Based on the message play information and the message push result, the preset feature extraction model and the preset message carousel prediction model are adjusted and optimized.

[0032] The present application also provides a message playback optimization device, which is a virtual device and is applied to a message playback optimization device. The message playback optimization device includes:

[0033] Establishing a module for acquiring group monitoring information, and inputting the group monitoring information into a preset feature extraction model to establish a group portrait corresponding to the group monitoring information;

[0034] A matching module, used to obtain message context features, and input the message context features and the group portrait into a preset message carousel prediction model, and match the message context features with the message carousel information corresponding to the group portrait;

[0035] The playing module is used to play the message according to the message carousel information.

[0036] Optionally, the matching module includes:

[0037] An acquisition unit, configured to acquire the message context features and the model input feature representation matrix corresponding to the group portrait;

[0038] The calculation unit is used to input the model input feature representation matrix into the preset message carousel prediction model, calculate the message carousel sequence that meets the preset target message revenue, and obtain the message carousel information.

[0039] Optionally, the computing unit includes:

[0040] A matching subunit, configured to input the model input feature representation matrix into the preset message carousel prediction model, and match the set of messages to be pushed corresponding to the model input feature representation matrix;

[0041] A calculation subunit, used to obtain one or more message carousel sequences corresponding to the message set to be pushed, and calculate the message revenue corresponding to each message carousel sequence;

[0042] The selection subunit is used to select a target message benefit that meets the preset target message benefit from each of the message benefits, and use the target message carousel sequence corresponding to the target message benefit as the message carousel information.

[0043] Optionally, the establishing module includes:

[0044] A first extraction unit, configured to input the group monitoring information into the user individual feature extraction model, extract the user individual features in the group monitoring information, and obtain the user individual feature recognition result;

[0045] The second extraction unit is used to input the user individual feature recognition result into the user group feature extraction model, extract the user group feature corresponding to the user individual feature recognition result, and obtain the group portrait.

[0046] Optionally, the first extraction unit includes:

[0047] A first extraction subunit is used to input the group monitoring information into the user coarse-grained feature recognition model, extract the user coarse-grained features of each of the target users, and obtain the user coarse-grained feature recognition result;

[0048] The second extraction subunit is used to input the group monitoring information into the user interaction mode recognition model, identify the user interaction mode between the target users, and obtain the user interaction mode recognition result.

[0049] Optionally, the second extraction unit includes:

[0050] A grouping subunit, configured to input the user individual feature recognition result into the user group feature extraction model, group each of the target users, and obtain the user group division result, wherein the user group division result includes one or more user groups;

[0051] An identification subunit, used to identify the social relationship of each user group and obtain the group social relationship identification result;

[0052] A subunit is established to identify the social relationship of each user group and obtain the group social relationship identification result.

[0053] Optionally, the message playback optimization device further includes:

[0054] A collection module, used to collect message playing information and message push results corresponding to the message playing information;

[0055] A feedback module is used to adjust and optimize the preset feature extraction model and the preset message carousel prediction model based on the message playback information and the message push result.

[0056] The present application also provides a message playback optimization device, which is a physical device. The message playback optimization device includes: a memory, a processor, and a program of the message playback optimization method stored in the memory and executable on the processor. When the program of the message playback optimization method is executed by the processor, the steps of the message playback optimization method as described above can be implemented.

[0057] The present application also provides a readable storage medium, on which is stored a program for implementing the message playback optimization method. When the program of the message playback optimization method is executed by a processor, the steps of the message playback optimization method as described above are implemented.

[0058] The present application obtains group monitoring information and inputs the group monitoring information into a preset feature extraction model to establish a group portrait corresponding to the group monitoring information, and then obtains message context features, and inputs the message context features and the group portrait into a preset message carousel prediction model, matches the message context features and the message carousel information corresponding to the group portrait, and then plays messages according to the message carousel information. That is, the present application provides a message playback optimization method based on group portrait, that is, by obtaining group monitoring information and inputting the group monitoring information into a preset feature extraction model, a group portrait corresponding to the group monitoring information is established, and then by obtaining message context features and inputting the message context features and the group portrait into a preset message carousel prediction model, the message context features and the message carousel information corresponding to the group portrait are matched, and then according to the message carousel information, targeted message playback is performed for the user group, thereby achieving the matching of the change status of the user group and the message carousel order, thereby avoiding the situation where the offline messages on the message screen do not conform to the preferences of the changed user group due to the change of the user group, thereby causing the offline message carousel efficiency to be low and the effect to be poor, thereby improving the playback efficiency and playback effect of the offline messages, thereby solving the technical problem of low efficiency and poor effect of the offline message carousel. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0060] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0061] Figure 1 This is a flow chart of the first embodiment of the message playback optimization method of the present application;

[0062] Figure 2 This is a flow chart of the second embodiment of the message playback optimization method of this application;

[0063] Figure 3This is a schematic diagram of adjusting and optimizing the message push in the message playback optimization method of this application;

[0064] Figure 4 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present application.

[0065] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0066] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0067] The present application embodiment provides a message playback optimization method, referring to Figure 1 In the first embodiment of the message playback optimization method of the present application, the message playback optimization method includes:

[0068] Step S10, obtaining group monitoring information, and inputting the group monitoring information into a preset feature extraction model to establish a group portrait corresponding to the group monitoring information;

[0069] In this embodiment, it should be noted that the message includes an advertisement, and the message playback optimization method can be applied to the field of advertising. The group monitoring information is information collected by sensors, wherein the sensors include visual sensors, sound sensors, infrared sensors, etc. Correspondingly, the group monitoring information includes visual information, sound information, infrared information, etc. The group portrait includes user individual characteristics and user group characteristics. The user individual characteristics are user individual characteristics of a single user, such as age, gender, and interaction methods with other users. The user group characteristics are user group characteristics of a user group, such as group classification label characteristics such as couples and families, group social relationship characteristics such as young people gatherings and parent-child travel, etc. The preset feature extraction model is a pre-trained machine learning model, wherein the preset feature extraction model includes a user individual feature recognition model and a user group feature recognition model.

[0070] Acquire group monitoring information, and input the group monitoring information into a preset feature extraction model to establish a group portrait corresponding to the group monitoring information. Specifically, collect the group monitoring information through a preset sensor, and input the group monitoring information into the user individual feature recognition model, so as to perform data processing on the feature representation matrix corresponding to the group monitoring information based on the data processing layer in the user individual feature recognition model, wherein the data processing layer includes a convolution layer, a pooling layer, a fully connected layer, etc., and the feature representation matrix is ​​a matrix storing the encoding corresponding to the group monitoring information. For example, assuming that the group monitoring information is visual information, the feature representation matrix is ​​a pixel matrix, and the pixel matrix includes one or more pixel encodings, thereby obtaining The user individual characteristics are obtained, wherein the user individual characteristics can be represented by a matrix or a vector. For example, assuming that a matrix is ​​used to represent the user individual characteristics, each column of the matrix corresponds to a user, and each row of the matrix corresponds to a characteristic. When the coding value in the matrix is ​​0, it indicates that the user does not have the corresponding characteristic. When the coding value in the matrix is ​​1, it indicates that the user has the corresponding characteristic. Further, the user individual characteristic representation matrix corresponding to the user individual characteristics is input into the user group characteristic recognition model, so as to perform data processing on the user individual characteristic representation matrix based on the data processing layer in the user group characteristic recognition model to obtain the user group characteristics, wherein the user group characteristics can be represented by a matrix or a vector.

[0071] Step S20, obtaining message context features, and inputting the message context features and the group portrait into a preset message carousel prediction model, matching the message context features and the message carousel information corresponding to the group portrait;

[0072] In this embodiment, it should be noted that the message context features are features extracted from the acquired message playback information, wherein the message playback information includes the type of the last played message, how many times a message has been played in the past 30 seconds, whether the target audience user group corresponding to a message has changed, etc. The group portrait is a group user portrait constructed based on the user group features, wherein the user group features include group division results and group social relationships, and the group portrait corresponds to one or more target users, each target user can be divided into one or more user groups. For example, there are a total of 4 target users in the elevator, and the group division results are that two target users are friend groups, and the other two target users are family groups. The corresponding group social relationships are young people traveling and parents and children traveling, respectively. Based on the group division results and the group social relationships, the group portrait corresponding to the friend group and the group portrait corresponding to the family group can be constructed respectively.

[0073] Acquire message context features, and input the message context features and the group portrait into a preset message carousel prediction model, match the message context features and the message carousel information corresponding to the group portrait, specifically, acquire message playback information, and extract features from the message playback information through a preset feature extraction model to obtain message context features, and then input the message context features and the group portrait into a preset message carousel prediction model, match the message context features and the data of items to be pushed corresponding to the group portrait, wherein the data of items to be pushed includes item click-through rates, and then sort the items to be pushed corresponding to each data of items to be pushed based on the item click-through rates in each data of items to be pushed, obtain the item click-through rate arrangement order of the data of items to be pushed, and then obtain the messages to be pushed corresponding to each data of items to be pushed, and use the item arrangement order as the playback order of each message to be pushed, that is, use the item arrangement order as the message carousel information corresponding to each message to be pushed.

[0074] In addition, the data of items to be pushed also includes actual revenue of items, and the actual revenue of items can be calculated based on the click rate of items and the revenue of items, and based on the actual revenue of each item, the items to be pushed corresponding to each data of items to be pushed are sorted to obtain the item revenue arrangement order of the data of items to be pushed, and then the messages to be pushed corresponding to each data of items to be pushed are obtained, and the item revenue arrangement order is used as the message carousel information of each message to be pushed. For example, assuming that the user group includes 2 user groups, wherein the item to be pushed corresponding to one user group is a couple hotel, the corresponding item click rate is 80%, and the average consumption of the couple hotel is 200, that is, the item revenue is 200, and the actual revenue of the item is 160, and the item to be pushed corresponding to the other user group is a parent-child restaurant, the corresponding item click rate is 50%, and the average consumption of the parent-child restaurant is 500, that is, the item revenue is 500, and then the actual revenue of the item is 250, and then the item revenue arrangement order is (parent-child restaurant, couple hotel), and then the message carousel information is to first play the message to be pushed corresponding to the parent-child restaurant, and then play the couple message corresponding to the couple hotel.

[0075] The step of inputting the message context feature and the group portrait into a preset message carousel prediction model and matching the message carousel information corresponding to the message context feature and the group portrait comprises:

[0076] Step S21, obtaining the message context feature and the model input feature representation matrix corresponding to the group portrait;

[0077] In this embodiment, it should be noted that the model input feature representation matrix includes one or more feature codes, and the feature code is an identifier of the message context feature, or an identifier of the individual user feature in the group portrait, or an identifier of the user group feature in the group portrait.

[0078] Step S22, inputting the model input feature representation matrix into the preset message carousel prediction model, calculating the message carousel sequence that meets the preset target message revenue, and obtaining the message carousel information.

[0079] In this embodiment, it should be noted that the preset target message revenue includes message click-through rate revenue, commodity consumption revenue corresponding to the message, etc., wherein, if the message click-through rate revenue is met, then when each message to be pushed is played based on the corresponding message carousel order, the sum of the probabilities of the user clicking on each message to be pushed is the highest; if the commodity consumption revenue is met, then after each message to be pushed is played based on the corresponding message carousel order, the total amount of consumption of the commodity corresponding to each message to be pushed by the user is the highest.

[0080] The model input feature representation matrix is ​​input into the preset message carousel prediction model, and the message carousel sequence that meets the preset target message revenue is calculated to obtain the message carousel information. Specifically, the model input feature representation matrix is ​​input into the preset message carousel prediction model, so as to perform data processing on the model input feature representation matrix based on the data processing layer in the preset message carousel prediction model to obtain the message carousel information vector, wherein the data processing layer includes a convolution layer, a pooling layer, a fully connected layer, etc., and then the message carousel information vector is obtained based on the preset target message revenue. Take the corresponding message carousel information. For example, assume that the message carousel information vector is (1, 2, 3, a, b, c), where 1, 2, 3, a, b, c are all message codes, and the message code is the identifier of the message to be pushed. If the preset target message revenue is click-through rate revenue, the message carousel information is (1, 2, 3), that is, the message carousel order is (1, 2, 3); if the preset target message revenue is commodity consumption revenue, the message carousel information is (a, b, c), that is, the message carousel order is (a, b, c).

[0081] The step of inputting the model input feature representation matrix into the preset message carousel prediction model, calculating the message carousel sequence that meets the preset target message revenue, and obtaining the message carousel information includes:

[0082] Step S221, inputting the model input feature representation matrix into the preset message carousel prediction model to match the set of messages to be pushed corresponding to the model input feature representation matrix;

[0083] In this embodiment, it should be noted that the set of messages to be pushed includes one or more messages to be pushed.

[0084] The model input feature representation matrix is ​​input into the preset message carousel prediction model to match the set of messages to be pushed corresponding to the model input feature representation matrix. Specifically, the model input feature representation matrix is ​​input into the preset message carousel prediction model to perform data processing on the model input feature representation matrix based on the data processing layer in the preset message carousel prediction model to obtain the encoding vector of messages to be pushed, that is, to obtain the set of messages to be pushed, wherein the encoding vector of messages to be pushed includes one or more message codes, wherein the message code is an identifier of the message to be pushed.

[0085] Step S222, obtaining one or more message carousel sequences corresponding to the message set to be pushed, and calculating the message revenue corresponding to each message carousel sequence;

[0086] In this embodiment, it should be noted that the message revenue includes click-through rate revenue and product charging revenue, etc.

[0087] Obtain one or more message carousel sequences corresponding to the set of messages to be pushed, and calculate the message revenue corresponding to each of the message carousel sequences. Specifically, sort the messages to be pushed in the set of messages to be pushed, obtain one or more message code arrangement sequences, and use the message code arrangement sequence as the message carousel sequence, and then calculate the message revenue corresponding to each of the message carousel sequences based on the historical message data corresponding to each of the messages to be pushed. For example, assuming that the message revenue is the click-through rate revenue, the set of messages to be pushed includes message A, message B and message C, and the message screen time is only Two messages can be played, and the message carousel order includes (A, B), (A, C), (B, A), (B, C), (C, A) and (C, B), and based on historical message data, the click-through rate of message A is 80%, the click-through rate of message B is 50%, and the click-through rate of message C is 40%. The click-through rate revenue corresponding to (A, B) is 0.8+0.5*x, wherein x is the coefficient of diminishing returns, that is, the messages ranked behind in the message carousel order, the more the message's revenue decreases, and the click-through rate revenue of each message carousel order can be calculated based on the diminishing returns coefficient.

[0088] Step S223, selecting a target message benefit that meets the preset target message benefit from each of the message benefits, and using the target message carousel sequence corresponding to the target message benefit as the message carousel information.

[0089] In this embodiment, it should be noted that the preset target message revenue is a message revenue threshold, wherein the message revenue threshold can be set to a maximum message revenue.

[0090] A target message benefit that meets the preset target message benefit is selected from each of the message benefits, and the target message carousel order corresponding to the target message benefit is used as the message carousel information. Specifically, a target message benefit that exceeds a message benefit threshold is selected from each of the message benefits, and based on the target message benefit, a target message carousel order corresponding to the target message benefit is selected from each of the message carousel orders as the message carousel information.

[0091] Step S30: Play messages according to the message carousel information.

[0092] In this embodiment, the message is played according to the message carousel information. Specifically, according to the message code arrangement order of the message carousel information, a message is selected from a preset message pool for carousel.

[0093] The step of playing the message according to the message carousel information includes:

[0094] Step S40, collecting message playing information and message push results corresponding to the message playing information;

[0095] In this embodiment, the message playback information includes message screen image information and sensor data, and the message push result corresponds to user consumption information, for example, whether the user has consumed at the merchant corresponding to the message to be pushed, the amount of consumption by the user at the merchant corresponding to the message to be pushed, etc.

[0096] Collect message playback information and the message push results corresponding to the message playback information, specifically, collect image data corresponding to the message playback information, wherein the image data includes message screen image information and visual sensor data, and convert the message screen image information and the visual sensor data into digital signal information, and then perform differential encryption processing on the digital signal information to obtain the message playback information, and collect user consumption information, and then perform differential encryption processing on the user consumption information to obtain the message push results.

[0097] Step S50: Based on the message play information and the message push result, the preset feature extraction model and the preset message carousel prediction model are adjusted and optimized.

[0098] In this embodiment, based on the message playback information and the message push result, the preset feature extraction model and the preset message carousel prediction model are adjusted and optimized. Specifically, the message playback information and the message push result are used as training data, and the preset feature extraction model and the preset message carousel prediction model are iteratively trained to optimize the preset feature extraction model and the preset message carousel prediction model.

[0099] This embodiment obtains group monitoring information and inputs the group monitoring information into a preset feature extraction model to establish a group portrait corresponding to the group monitoring information, and then obtains message context features, and inputs the message context features and the group portrait into a preset message carousel prediction model, matches the message context features and the message carousel information corresponding to the group portrait, and then plays messages according to the message carousel information. That is, the present embodiment provides a message playback optimization method based on group portrait, that is, by obtaining group monitoring information and inputting the group monitoring information into a preset feature extraction model, a group portrait corresponding to the group monitoring information is established, and then by obtaining message context features and inputting the message context features and the group portrait into a preset message carousel prediction model, the message context features and the message carousel information corresponding to the group portrait are matched, and then according to the message carousel information, targeted message playback is performed for the user group, thereby achieving the matching of the change status of the user group and the message carousel order, thereby avoiding the situation where the offline messages on the message screen do not conform to the preferences of the changed user group due to the change of the user group, thereby causing the offline message carousel efficiency to be low and the effect to be poor, thereby improving the playback efficiency and playback effect of the offline messages, and thus solving the technical problem of low efficiency and poor effect of the offline message carousel.

[0100] Further, refer to Figure 2 Based on the first embodiment of the present application, in another embodiment of the present application, the preset feature extraction model includes a user individual feature extraction model and a user group feature extraction model.

[0101] The step of acquiring the group monitoring information, inputting the group monitoring information into a preset feature extraction model, and establishing a group portrait corresponding to the group monitoring information includes:

[0102] Step S11, inputting the group monitoring information into the user individual feature extraction model, extracting the user individual features in the group monitoring information, and obtaining the user individual feature recognition result;

[0103] In this embodiment, it should be noted that the user individual feature extraction model is a pre-trained machine learning model.

[0104] The group monitoring information is input into the user individual feature extraction model, and the user individual features in the group monitoring information are extracted to obtain the user individual feature recognition result. Specifically, the group monitoring information is input into the user individual feature extraction model, and based on the data processing layer in the user individual feature extraction model, the data processing layer includes a convolution layer, a pooling layer, a fully connected layer, etc., the feature representation matrix corresponding to the group monitoring information is processed to obtain the user individual feature representation matrix, wherein each column of the user individual feature representation matrix corresponds to a user, and each column of the user individual feature representation matrix includes one or more feature codes, and each feature code corresponds to a user individual feature. For example, assuming that a column of the user individual feature representation matrix is ​​a vector (1, 3, 7), wherein feature code 1 indicates that the user is 10 years old, feature code 3 indicates that the user is male, and feature code 7 indicates that the way the user interacts with others is face-to-face dialogue.

[0105] The user individual feature extraction model includes a user coarse-grained feature recognition model and a user interaction mode recognition model, the user individual feature recognition result includes a user coarse-grained feature recognition result and a user interaction mode recognition result, and the user individual feature includes the user coarse-grained features and user interaction modes of one or more target users.

[0106] The step of inputting the group monitoring information into the user individual feature extraction model, extracting the user individual features in the group monitoring information, and obtaining the user individual feature recognition result comprises:

[0107] Step S111, inputting the group monitoring information into the user coarse-grained feature recognition model, extracting the user coarse-grained features of each of the target users, and obtaining the user coarse-grained feature recognition results;

[0108] In this embodiment, it should be noted that the coarse-grained features are individual features of users that do not include precise biometric information, wherein the precise biometric information includes face, fingerprint, etc. For example, the coarse-grained features may be gender, age, clothing, etc., thereby protecting the user's sensitive privacy information.

[0109] The group monitoring information is input into the user coarse-grained feature recognition model, and the coarse-grained features of each target user are extracted to obtain the user coarse-grained feature recognition result. Specifically, the group monitoring information is input into the user coarse-grained feature recognition model to perform a preset number of convolution and pooling alternating processes on the feature representation matrix corresponding to the group monitoring information to obtain the convolution and alternating processing results, and then the convolution and pooling processes are fully connected to obtain the user coarse-grained feature representation matrix, wherein each column of the user coarse-grained feature representation matrix corresponds to a user, and each column of the user coarse-grained feature representation matrix includes one or more feature codes, and each feature code corresponds to a user coarse-grained feature.

[0110] Step S112: input the group monitoring information into the user interaction mode recognition model, identify the user interaction mode between the target users, and obtain the user interaction mode recognition result.

[0111] In this embodiment, the interaction mode is an interaction behavior between users, for example, the interaction behavior includes face to face, back to back, hugging, holding hands, etc.

[0112] The group monitoring information is input into the user interaction mode recognition model to identify the interaction mode between each of the target users and obtain the user interaction mode recognition result. Specifically, the group monitoring information is input into the user interaction mode recognition model to perform a preset number of convolution and pooling alternating processes on the feature representation matrix corresponding to the group monitoring information to obtain the convolution and alternating processing results, and then the convolution and pooling processes are fully connected to obtain a user interaction mode representation matrix, wherein each column of the user interaction mode representation matrix corresponds to a user, each column of the user interaction mode representation matrix includes one or more feature codes, and each feature code corresponds to a user interaction mode.

[0113] Step S12: input the user individual feature recognition result into the user group feature extraction model, extract the user group feature corresponding to the user individual feature recognition result, and obtain the group portrait.

[0114] In this embodiment, it should be noted that the user group characteristics are group social relationships of the user group, such as couples, families, and the like.

[0115] The user individual feature recognition result is input into the user group feature extraction model, the user group feature corresponding to the user individual feature is extracted, and the group portrait is obtained. Specifically, the user individual feature recognition result is input into the user group feature extraction model, so as to perform data processing on the user individual feature representation matrix corresponding to the user individual feature recognition result based on the data processing layer of the user group feature extraction model, wherein the data processing layer includes a convolution layer, a pooling layer, a fully connected layer, etc., wherein the data processing includes convolution, pooling, full connection, etc., to obtain a user group feature representation matrix, and then construct the group portrait based on the user individual feature representation matrix and the user group feature representation matrix, wherein the group portrait corresponds to one or more users, and in the group portrait, each of the users corresponds to one or more user individual feature codes and one or more user group feature codes, wherein the user individual feature code is an identifier of the user individual feature, and the user group feature code is an identifier of the user group feature.

[0116] The user individual feature recognition result corresponds to one or more target users, and the user group feature includes the user group division result and the group social relationship recognition result.

[0117] The step of inputting the user individual feature recognition result into the user group feature extraction model, extracting the user group feature corresponding to the user individual feature recognition result, and obtaining the group portrait comprises:

[0118] Step S121, inputting the user individual feature recognition result into the user group feature extraction model, grouping each of the target users, and obtaining the user group division result, wherein the user group division result includes one or more user groups;

[0119] In this embodiment, the user individual feature recognition result is input into the user group feature extraction model, each of the target users is grouped, and the user group division result is obtained, wherein the user group division result includes one or more user groups. Specifically, the user individual feature recognition result is input into the user group feature extraction model to perform a preset number of convolution and pooling alternating processing on the user individual feature representation matrix corresponding to the user individual feature recognition result to obtain the convolution and pooling alternating processing result, and then the convolution and pooling alternating processing result is fully connected to obtain a user group feature representation matrix, and then the user grouping code is extracted from the user group feature representation matrix, and one or more user groups corresponding to the user grouping code are queried, wherein the user grouping code is the identifier of each of the user groups.

[0120] Step S122, identifying the social relationship of each user group, and obtaining the group social relationship identification result;

[0121] In this embodiment, the social relationships of each of the user groups are identified to obtain the group social relationship identification results. Specifically, the social relationship codes corresponding to each of the user groups are extracted from the user group feature representation matrix, and the group social relationship identification results are queried based on each of the social relationship codes, wherein the social relationship codes are identifiers of each of the social relationships.

[0122] Step S123: establishing the group portrait based on the user group division result and the group social relationship identification result.

[0123] In this embodiment, the group portrait is established based on the user group division result and the group social relationship identification result. Specifically, based on the user group division result and the group social relationship identification result, the group portrait is constructed in conjunction with the user individual feature representation matrix, wherein the group portrait can be represented by a matrix. For example, assuming that the group portrait is represented by a matrix, each column of the matrix corresponds to a user. Further, assuming that a column of the matrix is ​​(a, b, c), feature code a represents user individual features, feature code b represents the user group division result, and feature code c represents the group social relationship identification result. Figure 3 The figure shows a schematic diagram of the adjustment and optimization during message carousel in this embodiment, wherein the user individual feature recognition model is the user individual feature extraction model, the user group feature recognition model is the user group feature extraction model, the playback context feature is the message context feature, the message carousel model based on group portrait is the preset message carousel prediction model, and the product display and verification is the process of collecting the message playback information and the message push results corresponding to the explainable message.

[0124] This embodiment obtains the user individual feature recognition result by inputting the group monitoring information into the user individual feature extraction model, extracting the user individual features in the group monitoring information, and then inputting the user individual feature recognition result into the user group feature extraction model, extracting the user group features corresponding to the user individual features, and obtaining the group portrait. That is, the present embodiment provides a method for establishing a group portrait by identifying individual user characteristics, wherein the individual user characteristics include coarse-grained user characteristics and user interaction methods, that is, based on the group monitoring information, the individual user characteristics are extracted by the user individual characteristic extraction model to obtain the user individual characteristic recognition result, and then based on the user individual characteristic recognition result, the user group characteristics in the group monitoring information are extracted by the user group characteristic extraction model, and then based on the user individual characteristics and the user group characteristics, the group portrait is constructed, and then based on the group portrait, targeted message carousels can be performed for the user group, thereby avoiding the situation where the offline messages on the message screen do not conform to the preferences of the changed user group due to changes in the user group, thereby avoiding the situation where the offline message carousel efficiency is low and the effect is poor due to changes in the user group, thereby improving the playback efficiency and playback effect of the offline messages, and thus laying a foundation for solving the technical problem of low efficiency and poor effect of offline message carousel.

[0125] Reference Figure 4 , Figure 4 It is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present application.

[0126] like Figure 4 As shown, the message playback optimization device may include: a processor 1001, such as a CPU, a memory 1005, and a communication bus 1002. The communication bus 1002 is used to realize the connection and communication between the processor 1001 and the memory 1005. The memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may optionally be a storage device independent of the aforementioned processor 1001.

[0127] Optionally, the message playback optimization device may also include a rectangular user interface, a network interface, a camera, an RF (Radio Frequency) circuit, a sensor, an audio circuit, a WiFi module, etc. The rectangular user interface may include a display screen (Display), an input submodule such as a keyboard (Keyboard), and the optional rectangular user interface may also include a standard wired interface and a wireless interface. The network interface may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0128] Those skilled in the art will understand that Figure 4 The message playback optimization device structure shown in the figure does not constitute a limitation on the message playback optimization device, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0129] like Figure 4 As shown, the memory 1005 as a computer storage medium may include an operating system, a network communication module, and a message playback optimization program. The operating system is a program that manages and controls the hardware and software resources of the message playback optimization device, and supports the operation of the message playback optimization program and other software and / or programs. The network communication module is used to realize the communication between the components inside the memory 1005, and the communication with other hardware and software in the message playback optimization system.

[0130] exist Figure 4 In the message playback optimization device shown, the processor 1001 is used to execute the message playback optimization program stored in the memory 1005 to implement the steps of any of the message playback optimization methods described above.

[0131] The specific implementation of the message playback optimization device of the present application is basically the same as the embodiments of the above-mentioned message playback optimization method, and will not be repeated here.

[0132] The present application embodiment provides a message playback optimization device, the message playback optimization device comprising:

[0133] Establishing a module for acquiring group monitoring information, and inputting the group monitoring information into a preset feature extraction model to establish a group portrait corresponding to the group monitoring information;

[0134] A matching module, used to obtain message context features, and input the message context features and the group portrait into a preset message carousel prediction model, and match the message context features with the message carousel information corresponding to the group portrait;

[0135] The playing module is used to play the message according to the message carousel information.

[0136] Optionally, the matching module includes:

[0137] An acquisition unit, configured to acquire the message context features and the model input feature representation matrix corresponding to the group portrait;

[0138] The calculation unit is used to input the model input feature representation matrix into the preset message carousel prediction model, calculate the message carousel sequence that meets the preset target message revenue, and obtain the message carousel information.

[0139] Optionally, the computing unit includes:

[0140] A matching subunit, configured to input the model input feature representation matrix into the preset message carousel prediction model, and match the set of messages to be pushed corresponding to the model input feature representation matrix;

[0141] A calculation subunit, used to obtain one or more message carousel sequences corresponding to the message set to be pushed, and calculate the message revenue corresponding to each message carousel sequence;

[0142] The selection subunit is used to select a target message benefit that meets the preset target message benefit from each of the message benefits, and use the target message carousel sequence corresponding to the target message benefit as the message carousel information.

[0143] Optionally, the establishing module includes:

[0144] A first extraction unit, configured to input the group monitoring information into the user individual feature extraction model, extract the user individual features in the group monitoring information, and obtain the user individual feature recognition result;

[0145] The second extraction unit is used to input the user individual feature recognition result into the user group feature extraction model, extract the user group feature corresponding to the user individual feature recognition result, and obtain the group portrait.

[0146] Optionally, the first extraction unit includes:

[0147] A first extraction subunit is used to input the group monitoring information into the user coarse-grained feature recognition model, extract the user coarse-grained features of each of the target users, and obtain the user coarse-grained feature recognition result;

[0148] The second extraction subunit is used to input the group monitoring information into the user interaction mode recognition model, identify the user interaction mode between the target users, and obtain the user interaction mode recognition result.

[0149] Optionally, the second extraction unit includes:

[0150] A grouping subunit, configured to input the user individual feature recognition result into the user group feature extraction model, group each of the target users, and obtain the user group division result, wherein the user group division result includes one or more user groups;

[0151] An identification subunit, used to identify the social relationship of each user group and obtain the group social relationship identification result;

[0152] A subunit is established to identify the social relationship of each user group and obtain the group social relationship identification result.

[0153] Optionally, the message playback optimization device further includes:

[0154] A collection module, used to collect message playing information and message push results corresponding to the message playing information;

[0155] A feedback module is used to adjust and optimize the preset feature extraction model and the preset message carousel prediction model based on the message playback information and the message push result.

[0156] The specific implementation of the message playback optimization device of the present application is basically the same as the various embodiments of the above-mentioned message playback optimization method, and will not be repeated here.

[0157] An embodiment of the present application provides a readable storage medium, and the readable storage medium stores one or more programs, and the one or more programs can also be executed by one or more processors to implement the steps of any of the message playback optimization methods described above.

[0158] The specific implementation of the readable storage medium of the present application is basically the same as the embodiments of the above-mentioned message playback optimization method, and will not be repeated here.

[0159] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent processing scope of the present application.

Claims

1. A message playback optimization method, characterized in that: The message playback optimization method comprises: Acquire group monitoring information, and input the group monitoring information into a preset feature extraction model to establish a group portrait corresponding to the group monitoring information; Acquire message context features, and input the message context features and the group portrait into a preset message carousel prediction model to match the message carousel information corresponding to the message context features and the group portrait, wherein the message context features are features extracted from the acquired message playback information; Play messages according to the message carousel information; The step of playing the message according to the message carousel information includes: Collecting message playing information and message push results corresponding to the message playing information, wherein the message push results correspond to user consumption information; Based on the message play information and the message push result, adjusting and optimizing the preset feature extraction model and the preset message carousel prediction model; The preset feature extraction model includes a user individual feature extraction model and a user group feature extraction model. The step of acquiring the group monitoring information, inputting the group monitoring information into a preset feature extraction model, and establishing a group portrait corresponding to the group monitoring information includes: Inputting the group monitoring information into the user individual feature extraction model, extracting the user individual features in the group monitoring information, and obtaining the user individual feature recognition result; The user individual feature recognition result is input into the user group feature extraction model, the user group feature corresponding to the user individual feature recognition result is extracted, and the group portrait is obtained, wherein the user group feature includes the user group division result and the group social relationship recognition result.

2. The message playback optimization method according to claim 1, characterized in that: The step of inputting the message context features and the group portrait into a preset message carousel prediction model and matching the message carousel information corresponding to the message context features and the group portrait comprises: Obtain a model input feature representation matrix corresponding to the message context feature and the group portrait; The model input feature representation matrix is ​​input into the preset message carousel prediction model, the message carousel sequence that meets the preset target message revenue is calculated, and the message carousel information is obtained.

3. The message playback optimization method according to claim 2, characterized in that: The step of inputting the model input feature representation matrix into the preset message carousel prediction model, calculating the message carousel sequence that meets the preset target message revenue, and obtaining the message carousel information comprises: Inputting the model input feature representation matrix into the preset message carousel prediction model to match the set of messages to be pushed corresponding to the model input feature representation matrix; Obtain one or more message carousel sequences corresponding to the message set to be pushed, and calculate the message revenue corresponding to each message carousel sequence; A target message benefit that meets the preset target message benefit is selected from each of the message benefits, and a target message carousel sequence corresponding to the target message benefit is used as the message carousel information.

4. The message playback optimization method according to claim 1, characterized in that: The user individual feature extraction model includes a user coarse-grained feature recognition model and a user interaction mode recognition model, the user individual feature recognition result includes a user coarse-grained feature recognition result and a user interaction mode recognition result, and the user individual features include user coarse-grained features and user interaction modes of one or more target users, The step of inputting the group monitoring information into the user individual feature extraction model, extracting the user individual features in the group monitoring information, and obtaining the user individual feature recognition result comprises: Inputting the group monitoring information into the user coarse-grained feature recognition model, extracting the user coarse-grained features of each of the target users, and obtaining the user coarse-grained feature recognition results; The group monitoring information is input into the user interaction mode recognition model to recognize the user interaction mode between the target users, and obtain the user interaction mode recognition result.

5. The message playback optimization method according to claim 1, characterized in that: The user individual feature recognition result corresponds to one or more target users, The step of inputting the user individual feature recognition result into the user group feature extraction model, extracting the user group feature corresponding to the user individual feature recognition result, and obtaining the group portrait comprises: Inputting the user individual feature recognition result into the user group feature extraction model, grouping each of the target users, and obtaining the user group division result, wherein the user group division result includes one or more user groups; Identify the social relationship of each user group and obtain the group social relationship identification result; The group portrait is established based on the user group division results and the group social relationship identification results.

6. A message playback optimization device, characterized in that: The message playback optimization device comprises: Establishing a module for acquiring group monitoring information, and inputting the group monitoring information into a preset feature extraction model to establish a group portrait corresponding to the group monitoring information; A matching module, used to obtain message context features, and input the message context features and the group portrait into a preset message carousel prediction model to match the message carousel information corresponding to the message context features and the group portrait, wherein the message context features are features extracted from the obtained message playback information; A playing module, used for playing messages according to the message carousel information; The message playback optimization device also includes: A collection module, used to collect message playing information and message push results corresponding to the message playing information, wherein the message push results correspond to user consumption information; A feedback module, used to adjust and optimize the preset feature extraction model and the preset message carousel prediction model based on the message playback information and the message push result; The preset feature extraction model includes a user individual feature extraction model and a user group feature extraction model, and the establishment module is also used for: Inputting the group monitoring information into the user individual feature extraction model, extracting the user individual features in the group monitoring information, and obtaining the user individual feature recognition result; The user individual feature recognition result is input into the user group feature extraction model, the user group feature corresponding to the user individual feature recognition result is extracted, and the group portrait is obtained, wherein the user group feature includes the user group division result and the group social relationship recognition result.

7. A message playback optimization device, characterized in that: The message playback optimization device comprises: a memory, a processor, and a program stored in the memory for implementing the message playback optimization method. The memory is used to store a program for implementing a message playback optimization method; The processor is used to execute a program for implementing the message playback optimization method to implement the steps of the message playback optimization method as described in any one of claims 1 to 5.

8. A readable storage medium, characterized in that: The readable storage medium stores a program for implementing the message playback optimization method, and the program for implementing the message playback optimization method is executed by a processor to implement the steps of the message playback optimization method as described in any one of claims 1 to 5.

Images