Method and apparatus for determining recommended object, computer device and storage medium

Abstract

The application relates to a method and device for determining a recommended object, a computer device, a storage medium and a computer program product, and relates to the technical field of artificial intelligence. The method comprises the following steps: obtaining an emotion recognition result corresponding to a target user; the emotion recognition result is obtained by recognizing a face image collected when the target user browses a historical object recommendation page based on a pre-trained emotion recognition model; in the emotion recognition result corresponding to the target user, a target emotion recognition result meeting a preset positive emotion condition is determined, and a historical object recommendation page corresponding to the target emotion recognition result is taken as a target object recommendation page; a to-be-recommended object is determined based on a target object category contained in the target object recommendation page, and a to-be-displayed object recommendation page containing the to-be-recommended object is displayed.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, computer device, and storage medium for determining a recommended object. Background Technology

[0002] With the development of big data technology, a technology for determining recommendation objects has emerged. This technology determines the user's preferences based on the user's historical object data, and then determines the recommendation objects based on the user's preferences.

[0003] Traditional techniques for determining recommended objects involve inputting the user's basic information and historical object data into a pre-trained object recommendation model, which outputs the object categories of the recommended objects. Then, based on the object categories of the recommended objects, a search is performed in the recommended object library to obtain the objects to be recommended, and an object recommendation page containing the objects to be recommended is displayed.

[0004] However, current object recommendation technologies may use historical object data from a long time ago, which may not match the user's current needs, resulting in low accuracy in recommending target objects. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, apparatus, computer device, computer-readable storage medium, and computer program product for determining recommended objects that can improve the accuracy of object recommendations, in order to address the aforementioned technical problems.

[0006] Firstly, this application provides a method for determining recommended candidates. The method includes:

[0007] Obtain the emotion recognition result corresponding to the target user; the emotion recognition result is obtained by recognizing the face image collected when the target user browses the historical object recommendation page based on a pre-trained emotion recognition model;

[0008] Among the emotion recognition results corresponding to the target user, the target emotion recognition result that meets the preset positive emotion conditions is determined, and the historical object recommendation page corresponding to the target emotion recognition result is used as the target object recommendation page.

[0009] Based on the target object categories contained in the target object recommendation page, determine the objects to be recommended, and display the object recommendation page containing the objects to be recommended.

[0010] In one embodiment, the target object recommendation page is used to construct a target object recommendation page set; the step of determining the object to be recommended based on the target object categories contained in the target object recommendation page includes:

[0011] For each target object recommendation page in the target object recommendation page set, the recommendation weight of each target object recommendation page is determined according to the preset update weight allocation strategy, the page number of each target object recommendation page, and the page number of the object recommendation page to be displayed;

[0012] The target object category ratio of the target object recommendation page set is determined based on the target object category ratio corresponding to the object category of each target object recommendation page and the recommendation weight of each target object recommendation page.

[0013] Based on the target object category ratio and the preset number of update objects, the objects to be recommended are determined.

[0014] In one embodiment, the emotion recognition result is used to construct an emotion recognition result set; the step of determining a target emotion recognition result that meets preset positive emotion conditions from the emotion recognition results corresponding to the target user, and using the historical object recommendation page corresponding to the target emotion recognition result as the target object recommendation page includes:

[0015] If, in the set of emotion recognition results, there are a first preset number of consecutive emotion recognition results that are positive emotions, then the emotion recognition results in the set of emotion recognition results are taken as the target emotion recognition results of the target emotion recognition result set, and the historical object recommendation page corresponding to the target emotion recognition result is taken as the target object recommendation page; wherein, the order of each emotion recognition result in the set of emotion recognition results is the same as the order of the face images corresponding to the emotion recognition results in the face image set; the order of each face image in the face image set is determined based on the chronological order of the acquisition time corresponding to the face images.

[0016] In one embodiment, the method further includes:

[0017] If, in the emotion recognition results corresponding to the target user, there are a second consecutive preset number of emotions identified as negative emotions, a manual service prompt message containing the terminal location data of the recommended user is sent to the customer service terminal.

[0018] In one embodiment, the display of the object recommendation page containing the object to be recommended includes:

[0019] The initial recommended object in the recommended object page is updated based on the object to be recommended, and the updated recommended object page containing the object to be recommended is displayed; the initial recommended object is determined based on the target user data of the target user based on a pre-trained object recommendation model.

[0020] In one embodiment, the pre-trained object recommendation model includes at least one first-level object recommendation sub-model and a second-level object recommendation sub-model; the step of updating the initial recommended objects in the object recommendation page based on the object to be recommended further includes:

[0021] The set of combined model identifiers is determined based on the preset generation strategy of the combined model identifiers and the preset number of identifiers;

[0022] For each combined model identifier in the combined model identifier set, based on the preset correspondence between identifier bits and primary object recommendation sub-models, the combined model corresponding to the combined model identifier is determined in the primary object recommendation sub-model set; the combined model includes the primary object recommendation sub-models corresponding to each identifier bit of the combined model identifier.

[0023] Based on the combined model and the secondary object recommendation sub-model, determine the initial object recommendation model;

[0024] The initial object recommendation models are validated based on the sample user dataset to determine the fitness value of each initial object recommendation model.

[0025] Based on the preset genetic algorithm, the fitness values, the combined model identifier set, and the preset iteration stopping condition, the initial object recommendation model corresponding to the highest fitness value is determined, and the pre-trained object recommendation model is obtained.

[0026] In one embodiment, before obtaining the emotion recognition result corresponding to the target user, the method further includes:

[0027] In response to an object recommendation command, the current object recommendation page is displayed, and the facial image of the target user is captured;

[0028] The emotion recognition model is used to perform emotion recognition on the collected facial images to obtain the emotion recognition result corresponding to the facial image of the target user.

[0029] In one embodiment, the pre-trained emotion recognition model includes at least one first-level emotion recognition sub-model and a second-level emotion recognition sub-model; the emotion recognition of the acquired facial images based on the pre-trained emotion recognition model to obtain the emotion recognition result corresponding to the target user's facial image includes:

[0030] For each of the aforementioned face images, the face image is input into each of the aforementioned first-level emotion recognition sub-models to obtain the initial emotion recognition result corresponding to each of the aforementioned first-level emotion recognition sub-models;

[0031] The initial emotion recognition results are input into the secondary emotion recognition sub-model to obtain the emotion recognition result corresponding to the face image.

[0032] In one embodiment, the pre-trained object recommendation model includes at least one first-level object recommendation sub-model and a second-level object recommendation sub-model; the response to the object recommendation instruction further includes:

[0033] The target user data is input into each of the first-level object recommendation sub-models to obtain the object category of the first-level recommended object corresponding to each of the first-level object recommendation sub-models; the object category of each of the first-level recommended objects is input into the second-level object recommendation sub-model to obtain the object category of the initial recommended object;

[0034] Based on the object category of the initial recommended object, the initial recommended object for each of the object recommendation pages to be displayed is determined.

[0035] Secondly, this application also provides a device for determining recommended objects. The device includes:

[0036] The acquisition module is used to acquire the emotion recognition result corresponding to the target user; the emotion recognition result is obtained by recognizing the face image collected when the target user browses the historical object recommendation page based on a pre-trained emotion recognition model;

[0037] The first determining module is used to determine the target emotion recognition result that meets the preset positive emotion conditions from the emotion recognition results corresponding to the target user, and to use the historical object recommendation page corresponding to the target emotion recognition result as the target object recommendation page.

[0038] The second determining module is used to determine the object to be recommended based on the target object category contained in the target object recommendation page, and to display the object to be recommended page containing the object to be recommended.

[0039] In one embodiment, the target object recommendation page is used to construct a target object recommendation page set; the second determining module is specifically used for:

[0040] For each target object recommendation page in the target object recommendation page set, the recommendation weight of each target object recommendation page is determined according to the preset update weight allocation strategy, the page number of each target object recommendation page, and the page number of the object recommendation page to be displayed;

[0041] The target object category ratio of the target object recommendation page set is determined based on the target object category ratio corresponding to the object category of each target object recommendation page and the recommendation weight of each target object recommendation page.

[0042] Based on the target object category ratio and the preset number of update objects, the objects to be recommended are determined.

[0043] In one embodiment, the emotion recognition result is used to construct an emotion recognition result set; the first determining module is specifically used for:

[0044] If, in the set of emotion recognition results, there are a first preset number of consecutive emotion recognition results that are positive emotions, then the emotion recognition results in the set of emotion recognition results are taken as the target emotion recognition results of the target emotion recognition result set, and the historical object recommendation page corresponding to the target emotion recognition result is taken as the target object recommendation page; wherein, the order of each emotion recognition result in the set of emotion recognition results is the same as the order of the face images corresponding to the emotion recognition results in the face image set; the order of each face image in the face image set is determined based on the chronological order of the acquisition time corresponding to the face images.

[0045] In one embodiment, the device for determining the recommended object further includes:

[0046] The sending module is used to send a human service prompt message containing the terminal location data of the recommended user to the customer service terminal when there are two consecutive preset number of negative emotions identified in the emotion recognition results corresponding to the target user.

[0047] In one embodiment, the second determining module is specifically used for:

[0048] The initial recommended object in the recommended object page is updated based on the object to be recommended, and the updated recommended object page containing the object to be recommended is displayed; the initial recommended object is determined based on the target user data of the target user based on a pre-trained object recommendation model.

[0049] In one embodiment, the pre-trained object recommendation model includes at least one first-level object recommendation sub-model and a second-level object recommendation sub-model; the device for determining the recommended object further includes:

[0050] The third determining module is used to determine the set of combined model identifiers based on the preset generation strategy of the combined model identifiers and the preset number of identifiers.

[0051] The fourth determining module is used to determine the combined model corresponding to each combined model identifier in the set of combined model identifiers, based on a preset correspondence between identifier bits and primary object recommendation sub-models; the combined model includes the primary object recommendation sub-models corresponding to each identifier bit of the combined model identifier.

[0052] The fifth determining module is used to determine the initial object recommendation model based on the combined model and the secondary object recommendation sub-model;

[0053] The sixth determining module is used to perform validation processing on each of the initial object recommendation models based on the sample user dataset, and to determine the fitness value of each of the initial object recommendation models;

[0054] The seventh determining module is used to determine the initial object recommendation model corresponding to the highest fitness value based on the preset genetic algorithm, each fitness value, the combined model identifier set, and the preset iteration stopping condition, so as to obtain the pre-trained object recommendation model.

[0055] In one embodiment, the means for determining the recommended object further includes:

[0056] The response module is specifically used to respond to an object recommendation instruction, display the current object recommendation page, and collect the facial image of the target user;

[0057] The emotion recognition module is used to perform emotion recognition on the acquired facial images based on a pre-trained emotion recognition model, and obtain the emotion recognition result corresponding to the facial image of the target user.

[0058] In one embodiment, the pre-trained emotion recognition model includes at least one first-level emotion recognition sub-model and a second-level emotion recognition sub-model; the emotion recognition module is specifically used for:

[0059] For each of the aforementioned face images, the face image is input into each of the aforementioned first-level emotion recognition sub-models to obtain the initial emotion recognition result corresponding to each of the aforementioned first-level emotion recognition sub-models;

[0060] The initial emotion recognition results are input into the secondary emotion recognition sub-model to obtain the emotion recognition result corresponding to the face image.

[0061] In one embodiment, the pre-trained object recommendation model includes at least one first-level object recommendation sub-model and a second-level object recommendation sub-model; the device for determining the recommended object further includes:

[0062] The input module is used to input target user data into each of the first-level object recommendation sub-models to obtain the object category of the first-level recommended object corresponding to each of the first-level object recommendation sub-models; and to input the object category of each of the first-level recommended objects into the second-level object recommendation sub-models to obtain the object category of the initial recommended object.

[0063] The eighth determining module is used to determine the initial recommending object for each of the object recommendation pages to be displayed based on the object category of the initial recommending object.

[0064] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps described in the first aspect.

[0065] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps described in the first aspect.

[0066] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps described in the first aspect.

[0067] The aforementioned method, apparatus, computer equipment, storage medium, and computer program product for determining recommended objects acquire the emotion recognition results corresponding to the target user. These emotion recognition results are obtained by recognizing facial images collected when the target user browses historical object recommendation pages, based on a pre-trained emotion recognition model. From the emotion recognition results corresponding to the target user, target emotion recognition results that meet preset positive emotion conditions are identified, and the historical object recommendation page corresponding to the target emotion recognition result is used as the target object recommendation page. Based on the target object categories contained in the target object recommendation page, objects to be recommended are determined, and a display page containing the objects to be recommended is shown. In this scheme, the facial images are collected when the target user browses historical object recommendation pages; therefore, the facial images can reflect the target user's facial expressions in real time, and the emotion recognition results determined based on the facial images can reflect the target user's emotions while browsing historical object recommendation pages. Since the emotion recognition results corresponding to the target object recommendation page meet preset positive emotion conditions, the target object recommendation page is a historical object recommendation page where the target user exhibits positive emotions. Therefore, determining objects to be recommended based on the target object recommendation page can improve the accuracy of object recommendations. Attached Figure Description

[0068] Figure 1 This is an application environment diagram of the method for determining recommended objects in one embodiment;

[0069] Figure 2 This is a flowchart illustrating a method for determining the object to be recommended in one embodiment;

[0070] Figure 3 This is a flowchart illustrating a method for determining a pre-trained object recommendation model in one embodiment.

[0071] Figure 4 This is a structural block diagram of a device for determining recommended objects in one embodiment;

[0072] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0073] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0074] In one embodiment, such as Figure 1 As shown, a method for determining recommendation objects is provided. This embodiment illustrates the application of this method to a terminal. It can be understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0075] Step 102: Obtain the emotion recognition results corresponding to the target user.

[0076] The emotion recognition result is obtained by recognizing facial images collected when the target user browses the historical object recommendation page based on a pre-trained emotion recognition model. The emotion recognition model includes at least one first-level emotion recognition sub-model and a second-level emotion recognition sub-model.

[0077] In this embodiment, the terminal acquires facial images captured when the target user browses the historical object recommendation page, and inputs each facial image into a pre-trained emotion recognition model, outputting the emotion recognition result corresponding to each facial image. In one embodiment, the first-level emotion recognition sub-model of the emotion recognition model includes a random forest model, a convolutional neural network, and a self-attention neural network; the second-level emotion recognition sub-model is a logistic model. In one embodiment, the logistic model is a binary classification logistic model. The emotion recognition result can include various emotions, such as positive emotions, negative emotions, and neutral emotions, etc. In one embodiment, the emotion recognition result is either a positive emotion or a negative emotion.

[0078] Step 104: Among the emotion recognition results corresponding to the target user, determine the target emotion recognition result that meets the preset positive emotion conditions, and use the historical object recommendation page corresponding to the target emotion recognition result as the target object recommendation page.

[0079] In this embodiment, the terminal matches the set of face images corresponding to each historical object recommendation page and the set of emotion recognition results corresponding to each historical object recommendation page based on the browsing time interval of each historical object recommendation page and the acquisition time of the face image to which the emotion recognition result belongs. Specifically, for each emotion recognition result, if the terminal determines that the acquisition time of the face image to which the emotion recognition result belongs belongs to a certain browsing time interval (for ease of distinction, this is called the matching browsing time interval), then the historical object recommendation page corresponding to the matching browsing time interval is taken as the historical object recommendation page corresponding to the emotion recognition result (for ease of distinction, this is called the matching historical object recommendation page). For each set of emotion recognition results corresponding to a historical object recommendation page, if the terminal determines that the emotion recognition results in the set of emotion recognition results meet the preset positive emotion conditions, then the set of emotion recognition results is taken as the target emotion recognition result set, and the historical object recommendation page corresponding to the target emotion recognition result set is taken as the target object recommendation page. The target emotion recognition result set includes multiple target emotion recognition results.

[0080] Step 106: Determine the objects to be recommended based on the target object categories contained in the target object recommendation page, and display the object recommendation page containing the objects to be recommended.

[0081] The target object recommendation page is used to construct the target object recommendation page set. The target object category is the object category in the target object recommendation page.

[0082] In this embodiment, for each target object recommendation page, the terminal counts the target object categories of the recommended objects included in the page, as well as the number of recommended objects in each category. Based on the target object categories and the number of recommended objects in each category, the terminal calculates the target object category ratio for each target object category in the recommendation page. For a set of target object recommendation pages, the terminal calculates the target object category ratio for each target object category based on the recommendation weight of each set and the target object category ratio. Based on the target object category ratio for each target object category and a preset number of updated objects, the terminal searches the recommendation object library for a preset number of objects to be recommended. The terminal updates the object recommendation page based on the preset number of objects to be recommended and displays the updated object recommendation page containing the objects to be recommended. Specifically, the object recommendation page contains initial recommended objects, which are determined based on a pre-trained object recommendation model. The terminal obtains the target user data of the target user and inputs the target user data into the pre-trained object recommendation model, outputting the object categories of the initial recommended objects. For each object recommendation page, the terminal searches the recommendation object library for the preset number of initial recommended objects based on the object category of the initial recommended object and the preset number of page recommended objects. The pre-trained object recommendation model includes at least one first-level object recommendation sub-model and a second-level object recommendation sub-model. The method for determining the first-level object recommendation sub-model is detailed in steps 302 to 310. In one embodiment, the second-level object recommendation sub-model is a Logistic model. The target user data includes the target user's basic information and historical purchase information. Basic information may include, but is not limited to, the target user's age and gender. Historical purchase information may include, but is not limited to, the historical object category and the historical purchase frequency of the historical purchase objects.

[0083] In the aforementioned method for determining recommended objects, the facial images are collected when the target user browses the historical object recommendation pages. Therefore, the facial images can reflect the target user's facial expressions in real time while browsing the historical object recommendation pages. Consequently, the emotion recognition results determined based on the facial images can reflect the target user's emotions while browsing the historical object recommendation pages. Since the emotion recognition results corresponding to the target object recommendation pages meet the preset positive emotion conditions, the target object recommendation pages are the object recommendation pages where the target user exhibits positive emotions. Therefore, determining the objects to be recommended based on the target object recommendation pages can improve the accuracy of object recommendations.

[0084] In one embodiment, such as Figure 2As shown, the target object recommendation page is used to construct the target object recommendation page set; the objects to be recommended are determined based on the target object categories contained in the target object recommendation page, including:

[0085] Step 202: For each target object recommendation page in the target object recommendation page set, determine the recommendation weight of each target object recommendation page according to the preset update weight allocation strategy, the page number of each target object recommendation page, and the page number of the object recommendation page to be displayed.

[0086] In this embodiment, for each target object recommendation page in the target object recommendation page set, the terminal calculates the difference between the page number of the object recommendation page to be displayed and the page number of the target object recommendation page to obtain the page number difference. The terminal calculates the recommendation weight of each target object recommendation page based on the page number difference corresponding to each target object recommendation page and a preset update weight allocation strategy. The sum of the recommendation weights of all target object recommendation pages in the target object recommendation page set is 1. It can be understood that the smaller the page number difference of a target object recommendation page, the smaller the recommendation weight of that target object recommendation page. In one embodiment, the values ​​of each recommendation weight can be constructed as an arithmetic sequence. In another embodiment, the recommendation weight of the target object recommendation page corresponding to the smallest page number difference is set as a preset first recommendation weight, and the sum of the recommendation weights of other target object recommendation pages is (1 - first recommendation weight), where the value of the first recommendation weight is a positive number and less than or equal to 1. This application does not limit the preset update weight allocation strategy; as long as the sum of the recommendation weights determined according to the preset update weight allocation strategy is 1, it falls within the protection scope of this application.

[0087] Step 204: Determine the target object category ratio of the target object recommendation page set based on the target object category ratio corresponding to the object category of each target object recommendation page and the recommendation weight of each target object recommendation page.

[0088] In this embodiment, for each target object recommendation page, the terminal calculates the target object category ratio (referred to as the pre-statistical ratio for ease of distinction) based on the object category ratio corresponding to the object category of the target object recommendation page and the recommendation weight of the target object recommendation page. Specifically, for each target object recommendation page, the terminal calculates the product of the object category ratio corresponding to the object category of the target object recommendation page and the recommendation weight of the target object recommendation page to obtain the target object category ratio of the target object recommendation page. The terminal calculates the pre-statistical ratios of each target object recommendation page to obtain the target object category ratio of the target object recommendation page set (referred to as the post-statistical ratio for ease of distinction). Specifically, in the target object recommendation page set, the terminal accumulates the pre-statistical ratios of the same target object category to obtain the post-statistical ratio corresponding to that target object category. Here, the target object category refers to the object category in the target object recommendation page.

[0089] Step 206: Determine the objects to be recommended based on the target object category ratio and the preset number of updated objects.

[0090] In this embodiment, for each target object category in the target object recommendation page set, the terminal calculates the product of the target object category ratio and the preset number of updated objects to obtain the number of objects to be recommended for that target object category. The sum of the target object category ratios of all target object categories in the target object recommendation page set is 1; the sum of the number of objects to be recommended for all target object categories in the target object recommendation page set is equal to the preset number of updated objects. For each target object category in the target object recommendation page set, the terminal searches for matching objects to be recommended in the recommendation object database based on the target object category and the number of objects to be recommended for that target object category. After performing the above processing (calculating the number of objects to be recommended for the target object category and searching for matching objects to be recommended for that target object category) on all target object categories in the target object recommendation page set, the terminal obtains the objects to be recommended corresponding to the target object recommendation page set. The preset number of updated objects is a positive integer and is less than or equal to the number of objects in the initial recommended objects in the object recommendation page to be displayed. The initial recommended objects are determined based on a pre-trained object recommendation model. It is understandable that the initial number of recommended objects on the recommended object page is the same as the number of objects that the recommended object page can display.

[0091] In this embodiment, the recommendation weight of the target object recommendation page is determined based on a preset update weight allocation strategy, the page number of each target object recommendation page, and the page number of the object to be displayed recommendation page. Then, based on the recommendation weight of the target object recommendation page, the proportion of target object categories on the target object recommendation page, and a preset number of updated objects, the objects to be recommended are determined. Since the recommendation weight in this scheme is related to the page number of the target object recommendation page and the page number of the object to be displayed recommendation page, and the page number relationship between the two is different for the same target object recommendation page but different object to be displayed recommendation pages, the recommendation weight of the target object recommendation page changes dynamically for different object to be displayed recommendation pages, thereby improving the recommendation accuracy of the objects to be recommended.

[0092] In one embodiment, the emotion recognition results are used to construct an emotion recognition result set; the emotion recognition results are used to construct an emotion recognition result set; among the emotion recognition results corresponding to the target user, the target emotion recognition result that meets the preset positive emotion conditions is determined, and the historical object recommendation page corresponding to the target emotion recognition result is used as the target object recommendation page, including:

[0093] If, in the set of emotion recognition results, there are a first number of consecutive preset emotion recognition results that are positive emotions, the emotion recognition results in the set of emotion recognition results will be used as the target emotion recognition results of the target emotion recognition result set, and the historical object recommendation page corresponding to the target emotion recognition result will be used as the target object recommendation page.

[0094] The order of each emotion recognition result in the emotion recognition result set is the same as the order of the corresponding face images in the face image set; the order of each face image in the face image set is determined based on the chronological order of the acquisition time of the face images.

[0095] In this embodiment, for each historical object recommendation page corresponding to an emotion recognition result set, the terminal counts the emotion recognition results of that emotion recognition result set. If the terminal counts a preset first number of consecutive emotion recognition results as positive emotions, then the terminal uses that emotion recognition result set as the target emotion recognition result set, uses the emotion recognition results in the target emotion recognition result set as the target emotion recognition result, and uses the historical object recommendation page corresponding to the target emotion recognition result set (or the target emotion recognition result) as the target object recommendation page. Here, one emotion recognition result set corresponds to one historical object recommendation page, and the acquisition time of the facial image to which each emotion recognition result belongs in the browsing time interval of the historical object recommendation page corresponding to that emotion recognition result set. In one embodiment, the first number is 5. For example, assuming that the emotion recognition result set 1 corresponding to historical object recommendation page 1 is {positive emotion, positive emotion, positive emotion, positive emotion, positive emotion, positive emotion, negative emotion, negative emotion}, for the emotion recognition result set 1 corresponding to historical object recommendation page 1, the terminal counts 6 consecutive positive emotions in the emotion recognition result set 1. If the terminal determines that the emotion recognition results in emotion recognition result set 1 meet the preset positive emotion condition that there are 5 consecutive positive emotion recognition results, the terminal will use the historical object recommendation page 1 as the target object recommendation page.

[0096] In this embodiment, by statistically analyzing the emotion recognition results of the emotion recognition result set corresponding to the historical object recommendation page, if the emotion recognition result set has a first preset number of consecutive positive emotion recognition results, then the terminal uses the historical object recommendation page corresponding to that emotion recognition result set as the target object recommendation page. In other words, the emotion recognition result set corresponding to the target object recommendation page has at least a first preset number of consecutive positive emotion recognition results. Therefore, the target recommended object on the target object recommendation page can be considered a recommended object that satisfies the target user, or a recommended object that matches the target user's needs. Thus, determining the object to be recommended based on such a target recommended object can improve the accuracy of object recommendations.

[0097] In one embodiment, the method further includes:

[0098] If, in the emotion recognition results corresponding to the target user, there are a second consecutive preset number of emotions identified as negative, a manual service prompt message containing the terminal location data of the recommended user will be sent to the customer service terminal.

[0099] In this embodiment, if a terminal (referred to as the "recommended target terminal" for ease of distinction) has a consecutive set second number of negative emotions identified in the emotion recognition results (i.e., the emotion recognition results corresponding to the target user) across all historical recommended target pages, a human service prompt is generated and sent to the customer service terminal. The human service prompt includes the location data of the recommended target terminal. Optionally, the set second number can be 20, 30, 31, 40, etc. In one embodiment, the location data can be the terminal serial number of the recommended target terminal. After receiving the human service prompt, the operator of the customer service terminal can directly go to the recommended target terminal and provide human service to the target user.

[0100] In this embodiment, if a predetermined second number of consecutive emotions are identified as negative in the emotion recognition results corresponding to all historical object recommendation pages, a human service prompt is generated. This provides human service to the target user, thereby reducing their negative emotions, and also improves the accuracy of object recommendations through human service.

[0101] In one embodiment, displaying a page recommending objects containing the objects to be recommended includes:

[0102] The initial recommended objects in the recommended object page are updated based on the objects to be recommended, and the updated recommended object page containing the objects to be recommended is then displayed.

[0103] The initial recommended objects are determined based on target user data of the target user using a pre-trained object recommendation model. The updated object recommendation page contains the target recommended objects, which include the objects to be recommended.

[0104] In this embodiment, for a page of recommended objects to be displayed that includes initial recommended objects, the terminal updates the preset number of initial recommended objects based on a preset number of updated objects to be recommended, thus obtaining the target recommended object. Optionally, the target recommended object can be the object to be recommended, or it can include both initial recommended objects and objects to be recommended. The preset number of updated objects is a positive integer and is less than or equal to the number of objects that the page of recommended objects to be displayed can display (or contain). In one embodiment, the preset number of updated objects is 50% of the number of objects that the page of recommended objects to be displayed can display (or contain). Specifically, for a page of recommended objects to be displayed that includes initial recommended objects, the terminal randomly selects a preset number of initial recommended objects (referred to as objects to be updated for ease of distinction) from the page of recommended objects to be displayed, and updates the objects to be updated based on the preset number of updated objects to be recommended, thus obtaining the target recommended object. The terminal then displays the updated page of recommended objects to be displayed that includes the target recommended object.

[0105] In this embodiment, an updated object recommendation page containing the target object is obtained by updating the initial recommended object in the object recommendation page based on the target object to be recommended. Since the target object recommendation page is a historical object recommendation page reflecting positive emotions of the target user, and the target object is determined based on the target object recommendation page after emotion recognition processing, the updated object recommendation page containing the target recommended object can improve the accuracy of object recommendations. The method for determining the initial recommended object in the object recommendation page is described in the following embodiment and will not be repeated here.

[0106] In one embodiment, such as Figure 3 As shown, the pre-trained object recommendation model includes at least one first-level object recommendation sub-model and a second-level object recommendation sub-model; before updating the initial recommended objects in the object recommendation page based on the object to be recommended, it also includes:

[0107] Step 302: Determine the set of combined model identifiers according to the preset generation strategy of combined model identifiers and the preset number of identifiers.

[0108] In this embodiment, for each identifier bit of the combined model identifier, the terminal generates an identifier for that identifier bit according to a preset combined model identifier generation strategy. After performing the above-described identifier generation process on each identifier bit, the terminal obtains a combined model identifier. After performing the above-described combined model identifier generation process a preset number of times, the terminal obtains a set of combined model identifiers. The set of combined model identifiers includes a preset number (L) of combined model identifiers, each combined model identifier having M identifier bits, and each identifier bit corresponding to a first-level object recommendation sub-model. L and M are both positive integers, where M is the number of first-level object recommendation sub-models.

[0109] Step 304: For each combined model identifier in the combined model identifier set, determine the combined model corresponding to the combined model identifier in the first-level object recommendation sub-model set according to the preset correspondence between the identifier bits and the first-level object recommendation sub-models.

[0110] The combined model includes primary object recommendation sub-models corresponding to each identifier bit of the combined model identifier. These primary object recommendation sub-models are pre-trained.

[0111] In this embodiment, for each combined model identifier in the combined model identifier set, the terminal matches the first-level object recommendation sub-model corresponding to each identifier in the first-level object recommendation sub-model set according to the preset correspondence between identifier bits and first-level object recommendation sub-models, thus obtaining the combined model corresponding to the combined model identifier. It can be understood that if there are L combined model identifiers, then there are also L combined models. The first-level object recommendation sub-model set includes M object recommendation sub-models. In one embodiment, M = 8. Specifically, the object recommendation sub-model set includes Random Forest, Extremely Randomized Trees, Adaptive Boosting (AdaBoost), Bootstrap Aggregating (Bagging), Gradient Boosting Decision Tree, Light Gradient Boosting Machine (lightgbm), Extreme Gradient Boosting (xgboost), and Catgorical Boost (catboost). The correspondence between the identifier and the first-level object recommendation sub-model can be represented by a correspondence table, as shown in Table 1 below.

[0112]

[0113] Table 1

[0114] In one embodiment, the combined model identifier is a binary code. If the identifier bit is 0, it indicates that the combined model corresponding to that identifier does not include the first-level object recommendation sub-model corresponding to that identifier bit; if the identifier bit is 1, it indicates that the combined model corresponding to that identifier includes the first-level object recommendation sub-model corresponding to that identifier bit. Assuming M=8 and the combined model identifier is 10100000, then the combined model corresponding to the combined model identifier 10100000 includes first-level object recommendation sub-model 1 and first-level object recommendation sub-model 3.

[0115] Step 306: Determine the initial object recommendation model based on the combined model and the secondary object recommendation sub-model.

[0116] The initial object recommendation model comprises a combined model and a second-level object recommendation sub-model. The combined model is a first-level object recommendation sub-model of the initial object recommendation model. The second-level object recommendation sub-model is a pre-trained second-level object recommendation sub-model.

[0117] In this embodiment, for each combined model, the terminal constructs an initial object recommendation model based on the combined model and the secondary object recommendation sub-model. In one embodiment, the secondary object recommendation sub-model is a Logistic model. Assuming the number of combined models L is 2, combined model 1 includes a primary object recommendation sub-model 1 and a primary object recommendation sub-model 3; combined model 2 includes a primary object recommendation sub-model 2 and a primary object recommendation sub-model 3. Therefore, the initial object recommendation model 1 corresponding to combined model 1 includes a primary object recommendation sub-model 1, a primary object recommendation sub-model 3, and a secondary object recommendation sub-model; the initial object recommendation model 1 corresponding to combined model 1 also includes a primary object recommendation sub-model 2, a primary object recommendation sub-model 3, and a secondary object recommendation sub-model.

[0118] Step 308: Validate each initial object recommendation model based on the sample user dataset to determine the fitness value of each initial object recommendation model.

[0119] The sample user dataset comprises subsets of sample user data corresponding to multiple sample users. Each subset includes multiple sample user data sets for a single sample user. The sample user data includes basic information about the sample user and their historical purchase information. Basic information may include, but is not limited to, the sample user's age and gender. Historical purchase information may include, but is not limited to, the historical purchase category and the number of times the historical purchase was made.

[0120] In this embodiment, the terminal uses cross-validation to validate each initial object recommendation model based on a sample user dataset, obtaining the predicted recommended object category for each initial object recommendation model. It can be understood that the number of predicted recommended object categories can be one or more. For each initial object recommendation model, the terminal calculates the fitness value of the initial object recommendation model based on the actual recommended object category corresponding to the sample user dataset and the predicted recommended object category of the initial object recommendation model. In one embodiment, the fitness is the Matthews correlation coefficient.

[0121] Step 310: Based on the preset genetic algorithm, fitness values, combined model identifier set, and preset iteration stopping conditions, determine the initial object recommendation model corresponding to the highest fitness value, and obtain the pre-trained object recommendation model.

[0122] In this embodiment, the terminal compares the fitness values ​​of each initial object recommendation model and selects the largest fitness values ​​(referred to as target fitness values ​​for easy distinction) from the fitness values ​​of each initial object recommendation model. The terminal constructs an initial population based on the initial object recommendation model to which the target fitness values ​​belong. The preset selection number X is a positive integer less than or equal to L. The terminal generates a second-generation population based on the initial population, the set of combined model identifiers corresponding to the combined models of the initial object recommendation models in the initial population (referred to as the initial population identifier set for easy distinction), and a preset genetic algorithm. Specifically, the terminal randomly selects two initial population identifiers from the initial population identifier set as first target initial population identifiers. The terminal randomly selects a identifier bit from each first target initial population identifier as a crossover identifier bit and randomly swaps the crossover identifier bits of the two first target initial population identifiers to obtain crossover-processed initial population identifiers. After performing crossover processing on the initial population identifier set for a preset number of crossover processes (Y), the terminal obtains 2Y crossover-processed initial population identifiers. Wherein, the number of crossover processes Y is an integer, and is less than or equal to the preset number of identifiers L. The number of crossover identifier bits can be preset or randomly determined, but the number of crossover identifier bits for the two first target initial population identifiers is equal. It can be understood that the crossover identifier bits of different first target initial population identifiers can be the same or different. Which crossover identifier bit of a first target initial population identifier is to be crossover processed with which crossover identifier bit of another first target initial population identifier is random. For example, suppose the first target initial population identifier 1 is 11110000, the first target initial population identifier 2 is 11001001, and the number of crossover identifier bits is 2; the crossover identifier bits of the first target initial population identifier 1 (11110000) are the first bit (1) and the sixth bit (0); the crossover identifier bits of the first target initial population identifier 2 (11001001) are the first bit (1) and the eighth bit (1); if the terminal crossovers the first target initial population identifier 1 (11110000) with the first target initial population identifier 2 (11001001), the crossover identifier bits of the first target initial population identifier 1 (11110000) are the first bit (1) and the eighth bit (1); if the terminal crossovers the first target initial population identifier 1 (11110000) with the first target initial population identifier 2 (11001001), the crossover identifier bits of the first target initial population identifier 2 (110 ... The first bit (1) of the first target initial population identifier 1 (11001001) is cross-processed with the eighth bit (1) of the first target initial population identifier 2 (11001001), and the sixth bit (0) of the first target initial population identifier 1 (11110000) is cross-processed with the first bit (1) of the first target initial population identifier 2 (11001001), resulting in the cross-processed initial population identifier 1 (11110100) and the cross-processed initial population identifier 2 (01110100). The terminal calculates the product of the preset mutation probability and the number of initial population identifiers in the initial population identifier set to obtain the number of mutated identifiers (Z). Based on the number of mutated identifiers, the terminal randomly selects the number of mutated identifiers of the second target initial population identifier from the initial population identifier set.The terminal performs mutation processing on each second target initial population identifier to obtain mutated target initial population identifiers. Specifically, for each second target initial population identifier, the terminal randomly selects a random number of identifier bits of the second target initial population identifier for mutation processing to obtain mutated target initial population identifiers. It can be understood that the selection of identifier bits for mutation processing is random, and the number of identifier bits for mutation processing is also random; the mutated identifier bits of different second target initial population identifiers may be the same or different. In one embodiment, the mutation processing involves mutating 1s in the identifier bits to 0s and mutating 0s in the identifier bits to 1s. The number of mutated identifiers Z is an integer, and X + 2Y + Z = L (i.e., the preset selection number X + 2 times the number of crossover processes Y + the number of mutated identifiers Z = the preset number of identifiers L). For example, assuming the number of mutated identifiers Z = 2, the second target initial population identifier 1 is (11111111), and the second target initial population identifier 2 is (00000000). For the initial population identifier 1 of the second target (11111111), the terminal randomly selected the first, third, and fourth identifier bits as the mutation processing identifier bits, and performed mutation processing on the mutation processing identifier bits of the initial population identifier 1 of the second target (01001111). For the initial population identifier 2 of the second target (00000000), the terminal randomly selected the second identifier bit as the mutation processing identifier bit, and performed mutation processing on the mutation processing identifier bit of the initial population identifier 2 of the second target (01000000). The above examples are merely illustrative and do not constitute a limitation on the mutation processing method of this application.

[0123] The terminal uses the initial object recommendation model from the second-generation population as the initial object recommendation model in step 308, and returns to execute step 308 until a preset iteration stopping condition is met. The preset iteration stopping condition is the number of iterations; specifically, it is the generation of the Nth generation population, where N is a positive integer greater than 1. The terminal selects the initial object recommendation model corresponding to the highest fitness value in the Nth generation population as the pre-trained object recommendation model.

[0124] In this embodiment, a genetic algorithm is used to determine the initial object recommendation model corresponding to the highest fitness value, and this initial object recommendation model is used as the pre-trained object recommendation model. Since the processing in the genetic algorithm is random, determining the pre-trained object recommendation model based on the genetic algorithm is objective and can avoid subjective errors, thus improving the recommendation accuracy of the pre-trained object recommendation model.

[0125] In one embodiment, before obtaining the emotion recognition result corresponding to the target user, the method further includes:

[0126] In response to an object recommendation command, the current object recommendation page is displayed, and the facial image of the target user is captured. Based on a pre-trained emotion recognition model, emotion recognition is performed on the captured facial image to obtain the emotion recognition result corresponding to the target user's facial image.

[0127] In this embodiment, the terminal responds to an object recommendation instruction and displays a current object recommendation page containing the target recommended object. The object recommendation instruction is used to obtain the current object recommendation page. The target recommended object is the recommended object contained in the current object recommendation page. Optionally, the target recommended object can be an initial recommended object, a pending recommended object, or both. The initial recommended object is determined based on a pre-trained object recommendation model; the method for determining the pending recommended object refers to steps 102 to 106. Each acquired face image is input into a pre-trained emotion recognition model for emotion recognition, and the emotion recognition result corresponding to the target user's face image is output. In one embodiment, the first-level emotion recognition sub-model of the emotion recognition model includes a random forest model, a convolutional neural network, and a self-attention neural network; the second-level emotion recognition sub-model is a Logistic model. In one embodiment, the Logistic model is a binary classification Logistic model. The emotion recognition result can include various emotions, such as positive emotions, negative emotions, and ordinary emotions. In one embodiment, the emotion recognition result is either a positive emotion or a negative emotion. In one embodiment, the terminal can be a self-service teller machine. When the target user clicks the "Enter Recommended Object Page" button on the terminal, the terminal receives and responds to the object recommendation instruction. The terminal displays the current object recommendation page (i.e., the first object recommendation page) containing the target recommended object. It can be understood that the target recommended object in the first object recommendation page is the initial recommended object. If the terminal determines the first object recommendation page as the target object recommendation page, the terminal determines the object to be recommended for the second object recommendation page based on the first object recommendation page. Specifically, the method for determining the object to be recommended is described in step 106. It can be understood that the target recommended object in the second object recommendation page at this time includes both the initial recommended object and the object to be recommended, or the target recommended object is the object to be recommended. If the terminal does not determine the first object recommendation page as the target object recommendation page, then the target object to be recommended for the second object recommendation page is the initial recommended object. The method for determining the target recommended object for other object recommendation pages to be displayed is similar to this (the method for determining the target recommended object for the second object recommendation page), and will not be described again.

[0128] In this embodiment, in response to an object recommendation command, the current object recommendation page is displayed, and a facial image of the target user is captured. Based on the facial image and a pre-trained emotion recognition model, the emotion recognition result of the target user is determined. Therefore, the facial image is an image of the target user's face captured when the target user browses the current object recommendation page. In other words, the facial image can reflect the target user's facial expressions in real time while browsing the current object recommendation page. Consequently, the emotion recognition result determined based on the facial image can reflect the target user's emotions while browsing the current object recommendation page. Therefore, determining the target recommended objects for subsequent object recommendation pages (i.e., the pages to be displayed) based on the emotion recognition result can better match the target user's real-time needs, thereby improving the accuracy of object recommendations.

[0129] In one embodiment, the pre-trained emotion recognition model includes at least one first-level emotion recognition sub-model and a second-level emotion recognition sub-model; based on the pre-trained emotion recognition model, emotion recognition is performed on the acquired facial images to obtain the emotion recognition result corresponding to the target user's facial image, including:

[0130] For each face image, the face image is input into each first-level emotion recognition sub-model to obtain the initial emotion recognition result corresponding to each first-level emotion recognition sub-model; the initial emotion recognition result is input into the second-level emotion recognition sub-model to obtain the emotion recognition result corresponding to the face image.

[0131] Among them, the first-level emotion recognition sub-model is a pre-trained first-level emotion recognition sub-model; the second-level emotion recognition sub-model is a pre-trained second-level emotion recognition sub-model.

[0132] In this embodiment, for each face image, the terminal inputs the face image into each first-level emotion recognition sub-model in a pre-trained emotion recognition model, and outputs the initial emotion recognition result corresponding to each pre-trained first-level emotion recognition sub-model. Optionally, the initial emotion recognition result can be an emotion value or a specific emotion. In one embodiment, the initial emotion recognition result is an emotion value, where a larger emotion value indicates a more positive emotion. In another embodiment, the initial emotion recognition result is a specific emotion; for example, the initial emotion recognition result can be a positive emotion or a negative emotion. The terminal inputs each initial emotion recognition result into a second-level emotion recognition sub-model in a pre-trained emotion recognition model, and outputs the emotion recognition result corresponding to the face image. The emotion recognition result can include multiple emotions, such as positive emotions, negative emotions, and ordinary emotions. In one embodiment, the emotion recognition result is either a positive emotion or a negative emotion.

[0133] In this embodiment, preliminary emotion recognition is performed on facial images using pre-trained first-level emotion recognition sub-models to obtain initial emotion recognition results. Then, final emotion recognition is performed based on these initial results using pre-trained second-level emotion recognition sub-models. This allows for the fusion of multiple first-level emotion recognition sub-models for final emotion recognition, reducing errors from individual first-level emotion recognition sub-models and improving the accuracy of emotion recognition.

[0134] In one embodiment, the pre-trained object recommendation model includes at least one first-level object recommendation sub-model and a second-level object recommendation sub-model; in response to an object recommendation instruction, it also includes...

[0135] Input the target user data into each first-level object recommendation sub-model to obtain the object category of the first-level recommended object corresponding to each first-level object recommendation sub-model; input the object category of each first-level recommended object into the second-level object recommendation sub-model to obtain the object category of the initial recommended object; based on the object category of the initial recommended object, determine the initial recommended object for each object recommendation page to be displayed.

[0136] Among them, target user data is the user data of the target users.

[0137] In this embodiment, the terminal inputs target user data into each first-level object recommendation sub-model, outputs the object category of the first-level recommended object corresponding to each first-level object recommendation sub-model, and inputs the object category of each first-level recommended object into the second-level object recommendation sub-model, outputting the object category of the second-level recommended object. The object category of the second-level recommended object is used as the object category of the initial recommended object. The first-level object recommendation sub-model is a pre-trained first-level object recommendation sub-model; the second-level object recommendation sub-model is a pre-trained second-level object recommendation sub-model. For each object recommendation page to be displayed, the terminal searches the recommendation object database for the recommended object corresponding to the object category of the initial recommended object, based on the object category of the initial recommended object and the number of objects in the initial recommended object of the object recommendation page to be displayed. It can be understood that the object category of the second-level recommended object output by the second-level object recommendation sub-model can be one or more.

[0138] In this embodiment, preliminary recommendation predictions are made based on the target user data of the target user through each first-level object recommendation sub-model to obtain the object category of each first-level recommended object. Finally, a recommendation prediction is made based on the object category of each first-level recommended object using the second-level object recommendation sub-model. This allows for the fusion of multiple first-level object recommendation sub-models for recommendation prediction, reducing the error introduced by a single first-level object recommendation sub-model and improving the accuracy of the recommendation predictions of the pre-trained object recommendation model.

[0139] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0140] Based on the same inventive concept, this application also provides a device for determining recommended objects to implement the method for determining recommended objects described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more embodiments of the device for determining recommended objects provided below can be found in the limitations of the method for determining recommended objects described above, and will not be repeated here.

[0141] In one embodiment, such as Figure 4 As shown, a device for determining recommended objects is provided, the device comprising:

[0142] The acquisition module 402 is used to acquire the emotion recognition result corresponding to the target user; the emotion recognition result is obtained by recognizing the face image collected when the target user browses the historical object recommendation page based on the pre-trained emotion recognition model;

[0143] The first determining module 404 is used to determine the target emotion recognition result that meets the preset positive emotion conditions in the emotion recognition results corresponding to the target user, and to use the historical object recommendation page corresponding to the target emotion recognition result as the target object recommendation page.

[0144] The second determining module 406 is used to determine the object to be recommended based on the target object category contained in the target object recommendation page, and to display the object recommendation page containing the object to be recommended.

[0145] In one embodiment, the target object recommendation page is used to construct a target object recommendation page set; the second determining module 406 is specifically used for:

[0146] For each target object recommendation page in the target object recommendation page set, the recommendation weight of each target object recommendation page is determined according to the preset update weight allocation strategy, the page number of each target object recommendation page, and the page number of the object recommendation page to be displayed.

[0147] The proportion of target object categories in the target object recommendation page set is determined based on the proportion of target object categories corresponding to the object categories on each target object recommendation page and the recommendation weight of each target object recommendation page.

[0148] The objects to be recommended are determined based on the proportion of target object categories and the preset number of objects to be updated.

[0149] In one embodiment, the emotion recognition results are used to construct an emotion recognition result set; the first determining module 404 is specifically used for:

[0150] If, in the emotion recognition result set, there are a first preset number of consecutive emotion recognition results that are positive emotions, the emotion recognition results in the emotion recognition result set are taken as the target emotion recognition results of the target emotion recognition result set, and the historical object recommendation page corresponding to the target emotion recognition result is taken as the target object recommendation page; wherein, the order of each emotion recognition result in the emotion recognition result set is the same as the order of the face images corresponding to the emotion recognition results in the face image set; the order of each face image in the face image set is determined based on the chronological order of the acquisition time corresponding to the face images.

[0151] In one embodiment, the means for determining the recommended object further includes:

[0152] The sending module is used to send a manual service prompt message containing the terminal location data of the recommended user to the customer service terminal when there are a second preset number of consecutive negative emotions identified in the emotion recognition results corresponding to the target user.

[0153] In one embodiment, the second determining module 406 is specifically used for:

[0154] The initial recommended objects in the recommended object page are updated based on the objects to be recommended, and the updated recommended object page containing the objects to be recommended is displayed. The initial recommended objects are determined based on the target user data of the target user using a pre-trained object recommendation model.

[0155] In one embodiment, the pre-trained object recommendation model includes at least one first-level object recommendation sub-model and a second-level object recommendation sub-model; the device for determining the recommended object further includes:

[0156] The third determining module is used to determine the set of combined model identifiers based on the preset generation strategy of the combined model identifiers and the preset number of identifiers.

[0157] The fourth determination module is used to determine the combined model corresponding to each combined model identifier in the set of combined model identifiers, based on the preset correspondence between the identifier bits and the primary object recommendation sub-models; the combined model includes the primary object recommendation sub-models corresponding to each identifier bit of the combined model identifier.

[0158] The fifth determination module is used to determine the initial object recommendation model based on the combined model and the secondary object recommendation sub-model;

[0159] The sixth determination module is used to validate the recommendation models of each initial object based on the sample user dataset and determine the fitness value of each initial object recommendation model.

[0160] The seventh determination module is used to determine the initial object recommendation model corresponding to the highest fitness value based on the preset genetic algorithm, each fitness value, the combined model identifier set, and the preset iteration stopping condition, so as to obtain the pre-trained object recommendation model.

[0161] In one embodiment, the means for determining the recommended object further includes:

[0162] The response module is specifically used to respond to object recommendation instructions, display the current object recommendation page, and collect the facial image of the target user;

[0163] The emotion recognition module is used to perform emotion recognition on the collected facial images based on a pre-trained emotion recognition model, and obtain the emotion recognition result corresponding to the facial image of the target user.

[0164] In one embodiment, the pre-trained emotion recognition model includes at least one first-level emotion recognition sub-model and a second-level emotion recognition sub-model; the emotion recognition module is specifically used for:

[0165] For each face image, the face image is input into each first-level emotion recognition sub-model to obtain the initial emotion recognition result corresponding to each first-level emotion recognition sub-model;

[0166] The initial emotion recognition results are input into the secondary emotion recognition sub-model to obtain the emotion recognition results corresponding to the face image.

[0167] In one embodiment, the pre-trained object recommendation model includes at least one first-level object recommendation sub-model and a second-level object recommendation sub-model; the device for determining the recommended object further includes:

[0168] The input module is used to input target user data into each first-level object recommendation sub-model to obtain the object category of the first-level recommended object corresponding to each first-level object recommendation sub-model; and to input the object category of each first-level recommended object into the second-level object recommendation sub-model to obtain the object category of the initial recommended object.

[0169] The eighth determination module is used to determine the initial recommended object for each object recommendation page based on the object category of the initial recommended object.

[0170] Each module in the aforementioned device for determining the recommended objects can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0171] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a method for determining recommended objects. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0172] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0173] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0174] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0175] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0176] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0177] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0178] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0179] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for determining a recommended object, characterized by, The method includes: Obtain the emotion recognition result corresponding to the target user; the emotion recognition result is obtained by recognizing the face image collected when the target user browses the historical object recommendation page based on a pre-trained emotion recognition model; Among the emotion recognition results corresponding to the target user, the target emotion recognition result that meets the preset positive emotion conditions is determined, and the historical object recommendation page corresponding to the target emotion recognition result is used as the target object recommendation page. The objects to be recommended are determined based on the target object categories contained in the target object recommendation page. The initial recommended object in the recommended object page is updated based on the object to be recommended, and the updated recommended object page containing the object to be recommended is displayed; the initial recommended object is determined based on the target user data of the target user based on a pre-trained object recommendation model. The pre-trained object recommendation model includes at least one first-level object recommendation sub-model and a second-level object recommendation sub-model; the process of updating the initial recommended objects in the object recommendation page based on the object to be recommended also includes: The set of combined model identifiers is determined based on the preset generation strategy of the combined model identifiers and the preset number of identifiers; For each combined model identifier in the combined model identifier set, based on the preset correspondence between identifier bits and primary object recommendation sub-models, the combined model corresponding to the combined model identifier is determined in the primary object recommendation sub-model set; the combined model includes the primary object recommendation sub-models corresponding to each identifier bit of the combined model identifier. Based on the combined model and the secondary object recommendation sub-model, determine the initial object recommendation model; The initial object recommendation models are validated based on the sample user dataset to determine the fitness value of each initial object recommendation model. Based on the preset genetic algorithm, the fitness values, the combined model identifier set, and the preset iteration stopping condition, the initial object recommendation model corresponding to the highest fitness value is determined, and the pre-trained object recommendation model is obtained.

2. The method of claim 1, wherein, The target object recommendation page is used to construct a target object recommendation page set; the step of determining the object to be recommended based on the target object categories contained in the target object recommendation page includes: For each target object recommendation page in the target object recommendation page set, the recommendation weight of each target object recommendation page is determined according to the preset update weight allocation strategy, the page number of each target object recommendation page, and the page number of the object recommendation page to be displayed; The target object category ratio of the target object recommendation page set is determined based on the target object category ratio corresponding to the object category of each target object recommendation page and the recommendation weight of each target object recommendation page. Based on the target object category ratio and the preset number of update objects, the objects to be recommended are determined.

3. The method of claim 1, wherein, The emotion recognition results are used to construct an emotion recognition result set; the step of determining the target emotion recognition result that meets the preset positive emotion conditions from the emotion recognition results corresponding to the target user, and using the historical object recommendation page corresponding to the target emotion recognition result as the target object recommendation page includes: If, in the set of emotion recognition results, there are a first preset number of consecutive emotion recognition results that are positive emotions, then the emotion recognition results in the set of emotion recognition results are taken as the target emotion recognition results of the target emotion recognition result set, and the historical object recommendation page corresponding to the target emotion recognition result is taken as the target object recommendation page; wherein, the order of each emotion recognition result in the set of emotion recognition results is the same as the order of the face images corresponding to the emotion recognition results in the face image set; the order of each face image in the face image set is determined based on the chronological order of the acquisition time corresponding to the face images.

4. The method of claim 1, wherein, The method further includes: If, in the emotion recognition results corresponding to the target user, there are a second consecutive preset number of emotions identified as negative emotions, a manual service prompt message containing the terminal location data of the recommended user is sent to the customer service terminal.

5. The method of claim 1, wherein, Before obtaining the emotion recognition result corresponding to the target user, the process also includes: In response to an object recommendation command, the current object recommendation page is displayed, and the facial image of the target user is captured; The emotion recognition model is used to perform emotion recognition on the collected facial images to obtain the emotion recognition result corresponding to the facial image of the target user.

6. The method of claim 5, wherein, The pre-trained emotion recognition model includes at least one first-level emotion recognition sub-model and a second-level emotion recognition sub-model; the emotion recognition result obtained by performing emotion recognition on the acquired facial images based on the pre-trained emotion recognition model to obtain the emotion recognition result corresponding to the facial image of the target user includes: For each of the aforementioned face images, the face image is input into each of the aforementioned first-level emotion recognition sub-models to obtain the initial emotion recognition result corresponding to each of the aforementioned first-level emotion recognition sub-models; The initial emotion recognition results are input into the secondary emotion recognition sub-model to obtain the emotion recognition result corresponding to the face image.

7. A determination apparatus of a recommendation object, characterized by, The device includes: The acquisition module is used to acquire the emotion recognition result corresponding to the target user; the emotion recognition result is obtained by recognizing the face image collected when the target user browses the historical object recommendation page based on a pre-trained emotion recognition model; The first determining module is used to determine the target emotion recognition result that meets the preset positive emotion conditions from the emotion recognition results corresponding to the target user, and to use the historical object recommendation page corresponding to the target emotion recognition result as the target object recommendation page. The second determining module is used to determine the object to be recommended based on the target object categories contained in the target object recommendation page, update the initial recommended object in the object to be displayed recommendation page based on the object to be recommended, and display the updated object to be displayed recommendation page containing the object to be recommended; the initial recommended object is determined based on the target user data of the target user based on a pre-trained object recommendation model; The pre-trained object recommendation model includes at least one first-level object recommendation sub-model and a second-level object recommendation sub-model; the device further includes: The third determining module is used to determine the set of combined model identifiers based on the preset generation strategy of the combined model identifiers and the preset number of identifiers. The fourth determining module is used to determine the combined model corresponding to each combined model identifier in the set of combined model identifiers, based on a preset correspondence between identifier bits and primary object recommendation sub-models; the combined model includes the primary object recommendation sub-models corresponding to each identifier bit of the combined model identifier. The fifth determining module is used to determine the initial object recommendation model based on the combined model and the secondary object recommendation sub-model; The sixth determining module is used to perform validation processing on each of the initial object recommendation models based on the sample user dataset, and to determine the fitness value of each of the initial object recommendation models; The seventh determining module is used to determine the initial object recommendation model corresponding to the highest fitness value based on the preset genetic algorithm, each fitness value, the combined model identifier set, and the preset iteration stopping condition, so as to obtain the pre-trained object recommendation model.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, characterised in that, It includes a computer program that, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

Images