Theme skin recommendation method and device, equipment and medium

Abstract

The invention provides a theme skin recommendation method and device, equipment and a medium, and relates to the technical field of computers. In the application, firstly, skin description data corresponding to each candidate theme skin is acquired, and skin preference data corresponding to a target user is acquired; secondly, performing semantic mining on the skin description data, and outputting a skin description vector; secondly, performing semantic mining on the skin preference data, and outputting a skin preference vector; then, for each candidate subject skin, based on the skin description vector and the skin preference vector corresponding to the candidate subject skin, determining recommendation parameters corresponding to the candidate subject skin; and finally, based on the recommendation parameter corresponding to each candidate theme skin, determining a target theme skin matched with the target user from the plurality of candidate theme skins. Based on the content, the problem that in the prior art, the reliability of theme skin recommendation is relatively low can be solved.

Description

Technical Field

[0001] This application relates to the field of computer technology. Specifically, it relates to a method, device, equipment and medium for recommending theme skins. Background Art

[0002] The user interface is an important basis for the interaction between the application and the user. Among them, the application designer will design skins for the user interface. When designing the skin, while considering aesthetics, it is also necessary to take into account the layout of skin controls on the user interface and the reasonable use of colors. However, for different users, the application requirements or preferences for skins will be different. Therefore, corresponding recommendations need to be made for users.

[0003] However, currently, most existing theme skin recommendation systems rely on traditional rule engines or recommendation algorithms based on collaborative filtering. These methods usually can only recommend skins based on simple user behavior data, such as click history, purchase records or ratings, etc. The disadvantage of these recommendation methods is that they cannot fully explore the deep needs of users. For example, when processing data, most existing recommendation systems use a simple keyword matching method, ignoring the deep semantic information behind the skin. This method is likely to result in a relatively low matching degree of skin recommendations. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a method, device, equipment and medium for recommending theme skins to improve the relatively low reliability of theme skin recommendations in the existing technology.

[0005] To achieve the above purpose, this application adopts the following technical solutions: A method for recommending theme skins, including: Obtain the skin description data corresponding to each of the multiple candidate theme skins corresponding to the target software application, and obtain the skin preference data corresponding to the target user, where the skin description data includes at least data in two dimensions of text and image; Perform semantic mining on the skin description data and output the corresponding skin description vector; Perform semantic mining on the skin preference data and output the corresponding skin preference vector; For each of the candidate theme skins, based on the skin description vector corresponding to the candidate theme skin and the skin preference vector, determine the recommendation parameter corresponding to the candidate theme skin; Based on the recommendation parameter corresponding to each of the candidate theme skins, determine the target theme skin that matches the target user among the multiple candidate theme skins, and when the target user uses the target software application, recommend the target theme skin to the target user.

[0006] In a preferred option of the present application, in the above-mentioned theme skin recommendation method, the step of performing semantic mining on the skin description data and outputting the corresponding skin description vector includes: Performing semantic mining on the background image data in the skin description data through a first semantic mining model, and outputting a background image vector corresponding to the skin description data; Performing semantic mining on the skin description text in the skin description data through a second semantic mining model, and outputting a skin description text vector corresponding to the skin description data, where the skin description text is used to reflect information in at least one dimension among the color matching, layout method, and control style of the corresponding candidate theme skin; Taking the semantic information represented by the skin description text vector as auxiliary semantic information and integrating it into the semantic information represented by the background image vector to strengthen the semantics of the background image vector and form a corresponding skin description vector.

[0007] In a preferred option of the present application, in the above-mentioned theme skin recommendation method, the step of performing semantic mining on the background image data in the skin description data through a first semantic mining model and outputting a background image vector corresponding to the skin description data includes: Loading the background image data in the skin description data into a first semantic mining model, where the first semantic mining model belongs to a trained neural network model; Performing convolution processing on the background image data and outputting a corresponding background image convolution vector; Performing a first sliding window segmentation on the background image convolution vector to form a corresponding plurality of first image sliding window vectors, and based on an attention mechanism, associatively aggregating the plurality of first image sliding window vectors to form a first image aggregation vector; Performing a second sliding window segmentation on the background image convolution vector to form a corresponding plurality of second image sliding window vectors, and based on an attention mechanism, associatively aggregating the plurality of second image sliding window vectors to form a second image aggregation vector, where the second sliding window segmentation is different from the first sliding window segmentation at least in one dimension among size, step length, and path; Based on an attention mechanism, associatively aggregating the first image aggregation vector and the second image aggregation vector to form a target image aggregation vector; Performing a summation operation on the target image aggregation vector and the background image convolution vector to form a background image vector corresponding to the skin description data.

[0008] In a preferred option of the present application, in the above-mentioned theme skin recommendation method, the step of taking the semantic information represented by the skin description text vector as auxiliary semantic information and fusing it into the semantic information represented by the background image vector to achieve semantic enhancement of the background image vector and form a corresponding skin description vector includes: Extract the color local vector corresponding to the color combination, the layout local vector corresponding to the layout method, and the control local vector corresponding to the control style from the skin description text vector; Based on the color local vector and the background image vector, determine the first correlation parameter between the color combination and the background image, and based on the layout local vector and the background image vector, determine the second correlation parameter between the layout method and the background image, and further, based on the control local vector and the background image vector, determine the third correlation parameter between the control style and the background image; Fuse the first correlation parameter, the second correlation parameter, and the third correlation parameter to obtain the corresponding global correlation parameter; Based on the global correlation parameter, perform semantic adjustment on the background image vector to form a corresponding skin description vector.

[0009] In a preferred option of the present application, in the above-mentioned theme skin recommendation method, the step of performing semantic mining on the skin preference data and outputting the corresponding skin preference vector includes: Perform vector space mapping on the user basic data, user behavior data, and user preference data in the skin preference data respectively to form the corresponding user basic mapping vector, user behavior mapping vector, and user preference mapping vector, where the user basic data includes age, gender, occupation, the frequently used functions in the target software application, and the usage time of the target software application, the user behavior data includes historical skin selections, skin usage frequencies, and skin usage durations, and the user preference data includes preferred colors, fonts, and interface layouts; Based on the user basic mapping vector, the user behavior mapping vector, and the user preference mapping vector, determine the corresponding skin preference vector.

[0010] In a preferred option of the present application, in the above-mentioned theme skin recommendation method, the step of determining the corresponding skin preference vector based on the user basic mapping vector, the user behavior mapping vector, and the user preference mapping vector includes: Perform vector space conversion processing on the user basic mapping vector and the user preference mapping vector respectively to make them converted to the vector space where the user behavior mapping vector is located, forming the corresponding first space conversion vector and second space conversion vector; Extract semantic information related to the first spatial transformation vector from the user behavior mapping vector to obtain a corresponding first related semantic vector; Extract semantic information related to the second spatial transformation vector from the user behavior mapping vector to obtain a corresponding second related semantic vector; Perform a summation operation on the user behavior mapping vector, the first related semantic vector, and the second related semantic vector to obtain a corresponding skin preference vector.

[0011] In a preferred selection of the present application, in the above-mentioned theme skin recommendation method, the step of extracting semantic information related to the first spatial transformation vector from the user behavior mapping vector to obtain a corresponding first related semantic vector includes: Perform a dot product operation on the first spatial transformation vector and the transposed vector of the user behavior mapping vector, and output a dot product parameter distribution, where the size of the first spatial transformation vector is 1*n, the size of the user behavior mapping vector is 1*n, and the size of the dot product parameter distribution is n*n; Based on the dot product parameter distribution, perform a weighted summation calculation on the user behavior mapping vector and output a corresponding first related semantic vector, where the size of the first related semantic vector is 1*n.

[0012] The present application also provides a theme skin recommendation device, including: A skin data acquisition module, configured to acquire skin description data corresponding to each candidate theme skin among a plurality of candidate theme skins corresponding to a target software application, and acquire skin preference data corresponding to a target user, where the skin description data includes at least data in two dimensions of text and image; A description data mining module, configured to perform semantic mining on the skin description data and output a corresponding skin description vector; A preference data mining module, configured to perform semantic mining on the skin preference data and output a corresponding skin preference vector; A recommendation parameter determination module, configured to determine a recommendation parameter corresponding to each candidate theme skin based on the skin description vector corresponding to the candidate theme skin and the skin preference vector; A theme skin determination module, configured to determine a target theme skin adapted to the target user among the plurality of candidate theme skins based on the recommendation parameter corresponding to each candidate theme skin, and when the target user uses the target software application, recommend the target theme skin to the target user.

[0013] On the above basis, the present application also provides an electronic device, including: A memory for storing a computer program; A processor connected to the memory for executing the computer program stored in the memory to implement the above-mentioned theme skin recommendation method.

[0014] On this basis, the present application also provides a computer-readable storage medium, in which a computer program is stored, and when the computer program runs, it executes each step of the above-mentioned theme skin recommendation method.

[0015] For the theme skin recommendation method, device, equipment and medium provided by the present application, first, obtain the skin description data corresponding to each candidate theme skin, and obtain the skin preference data corresponding to the target user; secondly, perform semantic mining on the skin description data and output a skin description vector; then, perform semantic mining on the skin preference data and output a skin preference vector; after that, for each candidate theme skin, determine the recommendation parameter corresponding to the candidate theme skin based on the skin description vector and the skin preference vector corresponding to the candidate theme skin; finally, based on the recommendation parameter corresponding to each candidate theme skin, determine the target theme skin adapted to the target user among multiple candidate theme skins. Based on the above, since semantic mining can be performed on the skin description data and the skin preference data, the deep semantic information therein can be mined, realizing a reliable representation of the theme skin and the skin preference. Therefore, the reliability of the determined recommendation parameter can be guaranteed, so that the determined target theme skin has a high degree of adaptation to the target user, enabling highly reliable theme skin recommendation, thereby improving the problem of relatively low reliability of theme skin recommendation in the prior art. Therefore, the actual application needs of users can be met, and the stickiness of users to the skin and the application can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] To make the above objects, features and advantages of the present application more obvious and understandable, the following specifically gives preferred embodiments and cooperates with the attached drawings to make detailed descriptions as follows.

[0017] Figure 1 It is a structural block diagram of an electronic device provided by an embodiment of the present application.

[0018] Figure 2 It is a schematic flow chart of a theme skin recommendation method provided by an embodiment of the present application.

[0019] Figure 3 It is a schematic diagram of semantic mining provided by an embodiment of the present application.

[0020] Figure 4 It is a schematic diagram of semantic enhancement provided by an embodiment of the present application.

[0021] Figure 5A block diagram of the theme skin recommendation device provided by an embodiment of the present application. Detailed implementation manners

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part rather than all of the embodiments of the present application. Components of the embodiments of the present application described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations.

[0023] Therefore, the detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application claimed, but merely represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the scope of protection of the present application.

[0024] As Figure 1 shown, an embodiment of the present application provides an electronic device. Among them, the electronic device may include a memory, a processor, and a theme skin recommendation device.

[0025] Specifically, the memory and the processor are directly or indirectly electrically connected to achieve data transmission or interaction. For example, the memory and the processor may be electrically connected through one or more communication buses or signal lines. The theme skin recommendation device includes at least one software function module stored in the memory in the form of software or firmware. The processor is configured to execute the executable computer program stored in the memory, such as the software function module and computer program included in the theme skin recommendation device, to implement the theme skin recommendation method provided by the embodiment of the present application.

[0026] Optionally, the memory may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc.

[0027] Moreover, the processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), a System on Chip (SoC), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0028] It can be understood that Figure 1 the structure shown is only schematic, and the electronic device may further include more or fewer components than those shown Figure 1 herein, or have a configuration different from that shown Figure 1 For example, it may further include a communication unit for information interaction with other devices.

[0029] Combined with Figure 2 , an embodiment of the present application further provides a theme skin recommendation method applicable to the above-mentioned electronic device. Among them, the method steps defined by the process related to the theme skin recommendation method can be implemented by the electronic device. The following will Figure 2 elaborate in detail on the specific process shown.

[0030] Step S110, obtain the skin description data corresponding to each candidate theme skin among the multiple candidate theme skins corresponding to the target software application, and obtain the skin preference data corresponding to the target user.

[0031] In an embodiment of the present application, the electronic device can obtain the skin description data corresponding to each candidate theme skin among the multiple candidate theme skins corresponding to the target software application, and obtain the skin preference data corresponding to the target user (i.e., the factors that may affect the target user's selection of the theme skin). Among them, the skin description data includes at least two-dimensional data of text (such as for reflecting color matching, layout methods, and control styles, etc.) and images (such as background images, etc.).

[0032] Step S120, perform semantic mining on the skin description data and output the corresponding skin description vector.

[0033] In an embodiment of the present application, after obtaining the skin description data, the electronic device may perform semantic mining on the skin description data and output a corresponding skin description vector. It should be noted that for each of the multiple candidate theme skins, semantic mining can be performed to output a skin description vector corresponding to each candidate theme skin. That is to say, the potential semantic information in the skin description data is mined and characterized in the form of a vector. In this way, a corresponding skin description vector can be obtained.

[0034] Step S130: Perform semantic mining on the skin preference data and output a corresponding skin preference vector.

[0035] In an embodiment of the present application, after obtaining the skin preference data, the electronic device may perform semantic mining on the skin preference data and output a corresponding skin preference vector. That is to say, the potential semantic information in the skin preference data is mined and characterized in the form of a vector. In this way, a corresponding skin preference vector can be obtained.

[0036] Step S140: For each of the candidate theme skins, based on the skin description vector corresponding to the candidate theme skin and the skin preference vector, determine a recommendation parameter corresponding to the candidate theme skin.

[0037] In an embodiment of the present application, after mining the skin description vector and the skin preference vector, the electronic device may, for each of the candidate theme skins, based on the skin description vector corresponding to the candidate theme skin and the skin preference vector, determine a recommendation parameter corresponding to the candidate theme skin. Exemplarily, the similarity between the skin description vector and the skin preference vector (such as cosine similarity, etc.) can be calculated, and then this similarity can be used as the corresponding recommendation parameter. That is to say, the higher the similarity between the vectors, the higher the fitness and the more recommended.

[0038] Step S150: Based on the recommendation parameter corresponding to each of the candidate theme skins, determine a target theme skin that is suitable for the target user among the multiple candidate theme skins.

[0039] In the embodiment of the present application, after obtaining the recommendation parameters, the electronic device may determine a target theme skin adapted to the target user from the multiple candidate theme skins based on the recommendation parameters corresponding to each candidate theme skin. For example, a candidate theme skin with the maximum recommendation parameter may be determined as the target theme skin. Additionally, in some cases, if there are multiple candidate theme skins with the maximum recommendation parameter, one of them (which can be selected arbitrarily or based on certain rules) may be selected as the target theme skin, or they may all be used as the target theme skins and recommended to the target user, enabling the target user to make corresponding selections according to actual needs. Among them, when the target user uses the target software application, the target theme skin is recommended to the target user.

[0040] Based on the above content, since semantic mining can be performed on the skin description data and skin preference data, enabling the extraction of deep semantic information therein and achieving a reliable representation of the theme skin and skin preference, the reliability of the determined recommendation parameters can be ensured. As a result, the determined target theme skin has a high degree of suitability for the target user, enabling highly reliable theme skin recommendation, thereby improving the relatively low reliability problem of theme skin recommendation in the prior art. Therefore, the actual application needs of users can be met, and the user stickiness to the skin and application can be increased.

[0041] Regarding step S120, it should be noted that the specific method for semantic mining of the skin description data is not limited and can be selected according to actual needs.

[0042] For example, in an alternative embodiment, semantic mining (such as word embedding and convolution) can be separately performed on the text data and image data in the skin description data, and then the semantic vectors in the two dimensions obtained by mining are concatenated to obtain a skin description vector. For another example, in another alternative embodiment, in order to reliably represent the true semantics of the corresponding candidate theme skin by the mined skin description vector, step S120 described above may further include step S121, step S122, and step S123, and the specific content is as follows.

[0043] Step S121: Perform semantic mining on the background image data in the skin description data through a first semantic mining model, and output a background image vector corresponding to the skin description data.

[0044] In an embodiment of the present application, the background image data in the skin description data can be semantically mined through a first semantic mining model to output a background image vector corresponding to the skin description data. The background image data can be a solid-color picture or a landscape picture, etc.

[0045] Step S122: Semantically mine the skin description text in the skin description data through a second semantic mining model to output a skin description text vector corresponding to the skin description data.

[0046] In an embodiment of the present application, the skin description text in the skin description data can be semantically mined through a second semantic mining model to output a skin description text vector corresponding to the skin description data. Among them, the skin description text is used to reflect at least one dimension of information in the color matching (for example, background color: dark blue; text color: light gray; button color: bright blue), layout method (for example, top: navigation bar (including application name, icon); middle: content area (content in list or card style); bottom: navigation bar ("Home", "Message", "Settings" buttons)), and control style (for example, button shape: rounded rectangle; button states: normal state, hover state, click state) of the corresponding candidate theme skin. Exemplarily, the second semantic mining model can include a word embedding model (such as Word2Vec, etc., which represents each word as a dense vector). In this way, the skin description text can be processed by the word embedding model for word embedding to obtain the corresponding skin description text vector. For example, the skin description text can be segmented to obtain the corresponding multiple words, and then each word is embedded respectively to form the corresponding word embedding vector. Finally, the word embedding vectors of each word can be concatenated to form the corresponding skin description text vector. For example, for a rounded rectangle, it can be segmented into "rounded corner" and "rectangle", and the corresponding word embedding vectors can be: "rounded corner": [0.34, 0.67, 0.81, 0.26, 0.34, 0.79, 0.77, 0.39,......, 0.48]; "rectangle": [0.86, 0.84, 0.86, 0.90, 0.42, 0.26, 0.63, 0.42,......, 0.97]. <>

[0047] Step S123: Use the semantic information represented by the skin description text vector as auxiliary semantic information and fuse it into the semantic information represented by the background image vector to enhance the semantics of the background image vector and form a corresponding skin description vector.

[0048] In the embodiment of the present application, after the skin description text vector and the background image vector are mined, the semantic information represented by the skin description text vector can be used as auxiliary semantic information and integrated into the semantic information represented by the background image vector to enhance the semantics of the background image vector and form a corresponding skin description vector. In this way, the semantic information of the data in the two dimensions of text and image in the skin description data can be fully represented by the skin description vector.

[0049] It can be understood that in the above step S121, the specific method for semantic mining of the background image data in the skin description data is not limited. For example, in an alternative embodiment, considering that the background image data plays an important role in the visual expression of the theme skin, in order to fully mine the potential semantic information in the background image data, the above step S121 may further include step S121a, step S121b, step S121c, step S121d, step S121e, and step S121f. The specific content of each step is as follows (in combination with Figure 3 ).

[0050] Step S121a: Load the background image data in the skin description data into the first semantic mining model.

[0051] In the embodiment of the present application, the background image data in the skin description data can be loaded into the first semantic mining model. In this way, subsequent semantic mining can be performed in the first semantic mining model. The first semantic mining model belongs to a trained neural network model, that is, by learning corresponding samples and labels to obtain a corresponding mapping relationship. In addition, the second semantic mining model can also be learned together.

[0052] Step S121b: Perform convolution processing on the background image data and output a corresponding background image convolution vector.

[0053] In the embodiment of the present application, convolution processing can be performed on the background image data to output a corresponding background image convolution vector. That is to say, the first semantic mining model can include a convolution network layer (constituted by at least one convolution kernel). In this way, through this convolution network layer, convolution processing can be performed on the background image data to form a corresponding background image convolution vector.

[0054] Step S121c: Perform a first sliding window segmentation on the background image convolution vector to form a corresponding plurality of first image sliding window vectors, and based on the attention mechanism, associate and aggregate the plurality of first image sliding window vectors to form a first image aggregation vector.

[0055] In an embodiment of the present application, after obtaining the background image convolution vector, the background image convolution vector can be subjected to a first sliding window segmentation to form corresponding multiple first image sliding window vectors, and based on the attention mechanism, the multiple first image sliding window vectors are associated and aggregated to form a first image aggregation vector. For example, the background image convolution vector is [0.4, 0.7, 0.1, 0.6, 0.4, 0.9, 0.7, 0.9]. Through the first sliding window segmentation, multiple first image sliding window vectors can be obtained, which can be [0.4, 0.7, 0.1, 0.6, 0.4, 0.9], [0.7, 0.1, 0.6, 0.4, 0.9, 0.7], [0.1, 0.6, 0.4, 0.9, 0.7, 0.9] respectively. Then, based on the first first image sliding window vector, cross-attention processing can be performed on the second first image sliding window vector to obtain a first cross-attention vector, and based on the first cross-attention vector, cross-attention processing can be performed on the third first image sliding window vector to obtain a second cross-attention vector (in other embodiments, when including more first image sliding window vectors, this can be extended by analogy until cross-attention processing is performed on the last first image sliding window vector to obtain the last cross-attention vector). Then, the second cross-attention vector (i.e., the last cross-attention vector) can be concatenated with the background image convolution vector. Finally, the concatenated vector is compressed through pooling or other means to obtain a first image aggregation vector with the same size as the background image convolution vector. Based on this, the associated semantics between different regions in the background image data can be fully captured.

[0056] Step S121d: Perform a second sliding window segmentation on the background image convolution vector to form corresponding multiple second image sliding window vectors, and based on the attention mechanism, associate and aggregate the multiple second image sliding window vectors to form a second image aggregation vector.

[0057] In an embodiment of the present application, after obtaining the background image convolution vector, the background image convolution vector can be subjected to a second sliding window segmentation to form corresponding multiple second image sliding window vectors, and based on the attention mechanism, the multiple second image sliding window vectors are associated and aggregated to form a second image aggregation vector. The specific processing method can be the same as that in step S121c and will not be elaborated here one by one. Among them, the second sliding window segmentation is different from the first sliding window segmentation in at least one dimension of size, step length, and path, so as to capture different associated semantic information in the background image data.

[0058] Step S121e: Based on the attention mechanism, associate and aggregate the first image aggregation vector and the second image aggregation vector to form a target image aggregation vector.

[0059] In an embodiment of the present application, after obtaining the first image aggregation vector and the second image aggregation vector, based on the attention mechanism, the first image aggregation vector and the second image aggregation vector can be associated and aggregated to form a target image aggregation vector. Exemplarily, based on the first image aggregation vector, cross-attention processing can be performed on the second image aggregation vector to obtain a corresponding cross-attention vector, and this cross-attention vector can be used as the target image aggregation vector. In this way, through further associated aggregation, further mining and capture of associated semantic information can be achieved to improve the accuracy of capturing associated semantic information.

[0060] Step S121f: Perform a summation operation on the target image aggregation vector and the background image convolution vector to form a background image vector corresponding to the skin description data.

[0061] In an embodiment of the present application, the target image aggregation vector and the background image convolution vector can be subjected to a summation operation to form a background image vector corresponding to the skin description data, that is, to realize the joint representation of associated semantic information and global semantic information. In other embodiments, the target image aggregation vector and the background image convolution vector can be concatenated, and then the concatenated vector can be compressed by means such as pooling to obtain a background image vector with the same size as the background image convolution vector. In this way, through the background image vector, both the associated semantic information and the global semantic information in the background image data can be represented.

[0062] It can be understood that in the above step S123, the specific manner of realizing the semantic enhancement of the background image vector is not limited. For example, in an alternative embodiment, considering that the skin description text includes various different semantic information, such as color matching, layout style, and control style, based on this, in order to achieve high-precision semantic enhancement, the above step S123 can further include step S123a, step S123b, step S123c, and step S123d. The specific content of each step is as follows (reference can be made to Figure 4 the content shown).

[0063] Step S123a: Extract a color local vector corresponding to the color matching, a layout local vector corresponding to the layout style, and a control local vector corresponding to the control style from the skin description text vector.

[0064] In an embodiment of the present application, a color local vector corresponding to a color combination (i.e., a concatenated vector of word embedding vectors corresponding to each word in the text corresponding to the color combination), a layout local vector corresponding to a layout mode (i.e., a concatenated vector of word embedding vectors corresponding to each word in the text corresponding to the layout mode), and a control local vector corresponding to a control style (i.e., a concatenated vector of word embedding vectors corresponding to each word in the text corresponding to the control style) can be extracted from the skin description text vector.

[0065] Step S123b: Based on the color local vector and the background image vector, determine a first correlation parameter between the color combination and the background image, and based on the layout local vector and the background image vector, determine a second correlation parameter between the layout mode and the background image, and further, based on the control local vector and the background image vector, determine a third correlation parameter between the control style and the background image.

[0066] In an embodiment of the present application, after extracting the three local vectors, a first correlation parameter between the color combination and the background image can be determined based on the color local vector and the background image vector, a second correlation parameter between the layout mode and the background image can be determined based on the layout local vector and the background image vector, and a third correlation parameter between the control style and the background image can be determined based on the control local vector and the background image vector. Among them, the correlation parameter can be implemented based on an attention mechanism. For example, the attention score can be used as the correlation parameter. Specifically, the color local vector can be mapped to a query vector, and the background image vector can be mapped to a key vector. In this way, the dot product between the query vector and the transposed vector of the key vector can be calculated to obtain a corresponding dot product parameter matrix, that is, the first correlation parameter. The calculation methods of the second correlation parameter and the third correlation parameter can be the same as that of the first correlation parameter.

[0067] Step S123c: Fuse the first correlation parameter, the second correlation parameter, and the third correlation parameter to obtain a corresponding global correlation parameter.

[0068] In an embodiment of the present application, after obtaining the three correlation parameters, the first correlation parameter, the second correlation parameter, and the third correlation parameter can be fused to obtain a corresponding global correlation parameter. For example, the average value of the first correlation parameter, the second correlation parameter, and the third correlation parameter can be calculated to obtain the global correlation parameter.

[0069] Step S123d: Based on the global correlation parameter, perform semantic adjustment on the background image vector to form a corresponding skin description vector.

[0070] In an embodiment of the present application, after obtaining the global relevance parameter, the background image vector can be semantically adjusted based on the global relevance parameter to form a corresponding skin description vector. Exemplarily, the global relevance parameter can be determined based on an attention mechanism. Thus, when performing semantic adjustment, it can also be further implemented based on the attention mechanism. For example, the background image vector can be weighted and summed based on the global relevance parameter to obtain the corresponding skin description vector.

[0071] Regarding step S130, it should be noted that the specific manner of semantic mining of the skin preference data is not limited and can be selected accordingly according to actual needs.

[0072] For example, in an alternative embodiment, the skin preference data can be text data. Thus, corresponding word embedding processing can be performed through a word embedding model to obtain a corresponding skin preference vector to improve the efficiency of semantic mining.

[0073] Again, for example, in another alternative embodiment, to improve the accuracy of semantic mining, the above step S130 can further include step S131 and step S132, and the specific content of each step is described as follows.

[0074] Step S131: Perform vector space mapping on the user basic data, user behavior data, and user preference data in the skin preference data respectively to form corresponding user basic mapping vectors, user behavior mapping vectors, and user preference mapping vectors.

[0075] In an embodiment of the present application, the user basic data, user behavior data, and user preference data in the skin preference data can be respectively subjected to vector space mapping to form corresponding user basic mapping vectors, user behavior mapping vectors, and user preference mapping vectors. Among them, the user basic data includes age, gender, occupation, common functions in the target software application, and the usage time of the target software application. The user behavior data includes historical skin selections, skin usage frequencies, and skin usage durations. The user preference data includes preferred colors, fonts, and interface layouts. Thus, when the user basic data, the user behavior data, and the user preference data all belong to text data, word embedding processing can be performed through a word embedding model to achieve vector space mapping of each data and obtain corresponding mapping vectors.

[0076] Step S132: Determine a corresponding skin preference vector based on the user basic mapping vector, the user behavior mapping vector, and the user preference mapping vector.

[0077] In an embodiment of the present application, after obtaining the user base mapping vector, the user behavior mapping vector, and the user preference mapping vector, a corresponding skin preference vector may be determined based on the user base mapping vector, the user behavior mapping vector, and the user preference mapping vector. That is to say, the mapping vectors in three directions may be further fused.

[0078] Optionally, in step S132 above, the specific manner of determining the skin preference vector is not limited. For example, in an alternative embodiment, in order to effectively fuse the mapping vectors in three directions, step S132 above may further include step S132a, step S132b, step S132c, and step S132d. The specific content of each step is described as follows.

[0079] Step S132a: Perform vector space conversion processing on the user base mapping vector and the user preference mapping vector respectively, so as to convert them to the vector space where the user behavior mapping vector is located, and form corresponding first space conversion vector and second space conversion vector.

[0080] In an embodiment of the present application, the user base mapping vector and the user preference mapping vector may be respectively subjected to vector space conversion processing, so as to convert them to the vector space where the user behavior mapping vector is located, and form corresponding first space conversion vector and second space conversion vector. It should be noted that since the user behavior data includes historical skin selections, skin usage frequencies, and skin usage durations, that is, the semantic information represented by the user behavior mapping vector is directly related to the user's historical skin usage semantics, accordingly, the corresponding semantic information may be mined as the main semantic information. Based on this, in order to facilitate the extraction of subsequent relevant semantics, the user base mapping vector and the user preference mapping vector may be first converted to the vector space where the user behavior mapping vector is located. Specifically, it may be implemented through at least one conversion network layer (when multiple conversion network layers are included, the multiple conversion network layers may be cascaded, that is, the output of the previous one is used as the input of the next one). Each conversion network layer may include a conversion function, such as y = Ax + b, where x represents the input (such as the user base mapping vector), A represents the weight matrix, b represents the bias parameter, and y represents the output (such as the first space conversion vector).

[0081] Step S132b: Extract the semantic information related to the first space conversion vector from the user behavior mapping vector to obtain a corresponding first related semantic vector.

[0082] In an embodiment of the present application, after obtaining the first spatial transformation vector, semantic information related to the first spatial transformation vector can be extracted from the user behavior mapping vector to obtain a corresponding first related semantic vector.

[0083] Step S132c: Extract semantic information related to the second spatial transformation vector from the user behavior mapping vector to obtain a corresponding second related semantic vector.

[0084] In an embodiment of the present application, after obtaining the second spatial transformation vector, semantic information related to the second spatial transformation vector can be extracted from the user behavior mapping vector to obtain a corresponding second related semantic vector.

[0085] Step S132d: Perform a summation operation on the user behavior mapping vector, the first related semantic vector, and the second related semantic vector to obtain a corresponding skin preference vector.

[0086] In an embodiment of the present application, after obtaining the first related semantic vector and the second related semantic vector, a summation operation can be performed on the user behavior mapping vector, the first related semantic vector, and the second related semantic vector to obtain a corresponding skin preference vector. Based on this, the skin preference vector can not only represent the global semantic information in the user behavior mapping vector, but also importantly represent the semantic information in the user behavior mapping vector that is related to the user base mapping vector and the user preference mapping vector respectively, thereby improving its semantic representation ability. In this way, the purpose of using the semantic information in the user behavior data as the main part and the semantic information in the user base data and the user preference data as the auxiliary part can be achieved.

[0087] It can be understood that in the above step S132b, the specific method for extracting semantic information related to the first spatial transformation vector from the user behavior mapping vector is not limited. For example, in an alternative embodiment, in order to effectively capture the relevant semantic information, the above step S132b may further include the following content (the specific processing process of step S132c may be the same as the following processing process and will not be elaborated here): First, a dot product operation can be performed on the first spatial transformation vector and the transposed vector of the user behavior mapping vector to output a dot product parameter distribution (i.e., a matrix of a parameter distribution), where the size of the first spatial transformation vector is 1*n, the size of the user behavior mapping vector is 1*n, the size of the corresponding transposed vector is n*1, and the size of the dot product parameter distribution is n*n; Secondly, based on the dot product parameter distribution, a weighted sum calculation can be performed on the user behavior mapping vector to output a corresponding first relevant semantic vector, where the size of the first relevant semantic vector is 1*n.

[0088] Combined with Figure 5 , the embodiment of the present application further provides a theme skin recommendation device applicable to the above-mentioned electronic device. The theme skin recommendation device may include a skin data acquisition module, a description data mining module, a preference data mining module, a recommendation parameter determination module, and a theme skin determination module.

[0089] The skin data acquisition module is configured to acquire skin description data corresponding to each candidate theme skin of a target software application, and acquire skin preference data corresponding to a target user, where the skin description data includes at least data in two dimensions of text and image. In the embodiment of the present application, the skin data acquisition module can be used to execute Figure 2 the steps S110 shown, and the relevant content of the skin data acquisition module can be referred to the description of step S110 above.

[0090] The description data mining module is configured to perform semantic mining on the skin description data and output a corresponding skin description vector. In the embodiment of the present application, the description data mining module can be used to execute Figure 2 the steps S120 shown, and the relevant content of the description data mining module can be referred to the description of step S120 above.

[0091] The preference data mining module is configured to perform semantic mining on the skin preference data and output a corresponding skin preference vector. In the embodiment of the present application, the preference data mining module can be used to execute Figure 2 the steps S130 shown, and the relevant content of the preference data mining module can be referred to the description of step S130 above.

[0092] The recommendation parameter determination module is configured to determine a recommendation parameter corresponding to each candidate theme skin based on the skin description vector corresponding to the candidate theme skin and the skin preference vector. In the embodiment of the present application, the recommendation parameter determination module can be used to execute Figure 2 the steps S140 shown, and the relevant content of the recommendation parameter determination module can be referred to the description of step S140 above.

[0093] The above-mentioned theme skin determination module is used to determine a target theme skin that fits the target user from the multiple candidate theme skins based on the recommendation parameters corresponding to each candidate theme skin. When the target user uses the target software application, the target theme skin is recommended to the target user. In the embodiments of the present application, the theme skin determination module can be used to execute Figure 2 the steps S150 shown in the figure. For the relevant content of the theme skin determination module, reference can be made to the description of step S150 above.

[0094] In the embodiments of the present application, corresponding to the above-mentioned theme skin recommendation method applied to the electronic device, a computer-readable storage medium is also provided. A computer program is stored in the computer-readable storage medium, and when the computer program runs, it executes each step of the theme skin recommendation method.

[0095] Among them, the steps executed when the foregoing computer program runs will not be elaborated here one by one, and reference can be made to the explanation of the theme skin recommendation method above.

[0096] In summary, for the theme skin recommendation method, device, equipment and medium provided in the present application, first, obtain the skin description data corresponding to each candidate theme skin, and obtain the skin preference data corresponding to the target user; secondly, perform semantic mining on the skin description data to output a skin description vector; then, perform semantic mining on the skin preference data to output a skin preference vector; after that, for each candidate theme skin, determine the recommendation parameters corresponding to the candidate theme skin based on the skin description vector and the skin preference vector corresponding to the candidate theme skin; finally, based on the recommendation parameters corresponding to each candidate theme skin, determine a target theme skin that fits the target user from the multiple candidate theme skins. Based on the above content, since semantic mining can be performed on the skin description data and the skin preference data, the deep semantic information therein can be mined, realizing a reliable representation of the theme skin and the skin preference. Therefore, the reliability of the determined recommendation parameters can be guaranteed, so that the determined target theme skin has a high degree of fit with the target user, enabling highly reliable theme skin recommendation, thereby improving the problem of relatively low reliability of theme skin recommendation existing in the prior art. Therefore, the actual application needs of users can be met, and the stickiness of users to the skin and the application can be improved.

[0097] In several embodiments provided by the embodiments of the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of devices, methods, and computer program products according to multiple embodiments of the present application. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the module, program segment, or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that marked in the accompanying drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, as well as the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0098] In addition, in each embodiment of the present application, the various functional modules may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0099] If the described functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, an electronic device, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes. It should be noted that in this article, the term "including", "comprising", or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article, or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article, or device including the said element.

[0100] The above are only the preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. A method for recommending a theme skin, characterized in that, Including: Obtaining skin description data corresponding to each of multiple candidate theme skins for a target software application, and obtaining skin preference data corresponding to a target user, where the skin description data includes at least data in two dimensions of text and image; Performing semantic mining on the skin description data and outputting a corresponding skin description vector; Performing semantic mining on the skin preference data and outputting a corresponding skin preference vector; For each of the candidate theme skins, determining a recommended parameter corresponding to the candidate theme skin based on the skin description vector corresponding to the candidate theme skin and the skin preference vector; Based on the recommended parameter corresponding to each of the candidate theme skins, determining a target theme skin adapted to the target user from the multiple candidate theme skins, and when the target user uses the target software application, recommending the target theme skin to the target user.

2. The method for recommending a theme skin according to claim 1, wherein The step of performing semantic mining on the skin description data and outputting a corresponding skin description vector includes: Performing semantic mining on the background image data in the skin description data through a first semantic mining model and outputting a background image vector corresponding to the skin description data; Performing semantic mining on the skin description text in the skin description data through a second semantic mining model and outputting a skin description text vector corresponding to the skin description data, where the skin description text is used to reflect information in at least one dimension of color matching, layout mode, and control style of the corresponding candidate theme skin; Taking the semantic information represented by the skin description text vector as auxiliary semantic information and fusing it into the semantic information represented by the background image vector to achieve semantic enhancement of the background image vector and form a corresponding skin description vector.

3. The theme skin recommendation method according to claim 2, wherein The step of performing semantic mining on the background image data in the skin description data through a first semantic mining model and outputting a background image vector corresponding to the skin description data includes: Loading the background image data in the skin description data into a first semantic mining model, where the first semantic mining model belongs to a trained neural network model; Performing convolution processing on the background image data and outputting a corresponding background image convolution vector; Performing a first sliding window segmentation on the background image convolution vector to form a corresponding plurality of first image sliding window vectors, and based on an attention mechanism, associatively aggregating the plurality of first image sliding window vectors to form a first image aggregation vector; Performing a second sliding window segmentation on the background image convolution vector to form a corresponding plurality of second image sliding window vectors, and based on an attention mechanism, associatively aggregating the plurality of second image sliding window vectors to form a second image aggregation vector, where the second sliding window segmentation is different from the first sliding window segmentation at least in one dimension of size, step length, and path; Based on an attention mechanism, associatively aggregating the first image aggregation vector and the second image aggregation vector to form a target image aggregation vector; Perform a summation operation on the target image aggregation vector and the background image convolution vector to form a background image vector corresponding to the skin description data.

4. The theme skin recommendation method according to claim 2, wherein The step of using the semantic information represented by the skin description text vector as auxiliary semantic information and fusing it into the semantic information represented by the background image vector to achieve semantic enhancement of the background image vector and form a corresponding skin description vector includes: Extract a color local vector corresponding to the color combination, a layout local vector corresponding to the layout method, and a control local vector corresponding to the control style from the skin description text vector; Based on the color local vector and the background image vector, determine a first correlation parameter between the color combination and the background image, and based on the layout local vector and the background image vector, determine a second correlation parameter between the layout method and the background image, and, based on the control local vector and the background image vector, determine a third correlation parameter between the control style and the background image; Fuse the first correlation parameter, the second correlation parameter, and the third correlation parameter to obtain a corresponding global correlation parameter; Based on the global correlation parameter, perform semantic adjustment on the background image vector to form a corresponding skin description vector.

5. The theme skin recommendation method according to any one of claims 1-4, characterized in that The step of performing semantic mining on the skin preference data and outputting a corresponding skin preference vector includes: Perform vector space mapping on the user basic data, user behavior data, and user preference data in the skin preference data respectively to form corresponding user basic mapping vectors, user behavior mapping vectors, and user preference mapping vectors, where the user basic data includes age, gender, occupation, common functions in the target software application, and the usage time of the target software application, the user behavior data includes historical skin selections, skin usage frequencies, and skin usage durations, and the user preference data includes preferred colors, fonts, and interface layouts; Based on the user basic mapping vector, the user behavior mapping vector, and the user preference mapping vector, determine a corresponding skin preference vector.

6. The theme skin recommendation method according to claim 5, characterized in that, The step of determining a corresponding skin preference vector based on the user basic mapping vector, the user behavior mapping vector, and the user preference mapping vector includes: Perform vector space conversion processing on the user basic mapping vector and the user preference mapping vector respectively to make them converted to the vector space where the user behavior mapping vector is located, forming corresponding first space conversion vectors and second space conversion vectors; Extract semantic information related to the first space conversion vector from the user behavior mapping vector to obtain a corresponding first related semantic vector; Extract semantic information related to the second space conversion vector from the user behavior mapping vector to obtain a corresponding second related semantic vector; Perform a summation operation on the user behavior mapping vector, the first related semantic vector, and the second related semantic vector to obtain a corresponding skin preference vector.

7. The theme skin recommendation method according to claim 6, wherein The step of extracting semantic information related to the first spatial transformation vector from the user behavior mapping vector to obtain a corresponding first relevant semantic vector includes: Performing a dot product operation on the first spatial transformation vector and the transposed vector of the user behavior mapping vector, and outputting a dot product parameter distribution, where the size of the first spatial transformation vector is 1*n, the size of the user behavior mapping vector is 1*n, and the size of the dot product parameter distribution is n*n; Based on the dot product parameter distribution, performing a weighted summation calculation on the user behavior mapping vector, and outputting a corresponding first relevant semantic vector, where the size of the first relevant semantic vector is 1*n.

8. A theme skin recommendation device, characterized in that, including: A skin data acquisition module, configured to acquire skin description data corresponding to each of a plurality of candidate theme skins corresponding to a target software application, and acquire skin preference data corresponding to a target user, where the skin description data includes at least data in two dimensions of text and image; A description data mining module, configured to perform semantic mining on the skin description data and output a corresponding skin description vector; A preference data mining module, configured to perform semantic mining on the skin preference data and output a corresponding skin preference vector; A recommendation parameter determination module, configured to determine a recommendation parameter corresponding to each candidate theme skin based on the skin description vector corresponding to the candidate theme skin and the skin preference vector; A theme skin determination module, configured to determine a target theme skin adapted to the target user from the plurality of candidate theme skins based on the recommendation parameter corresponding to each candidate theme skin, and recommend the target theme skin to the target user when the target user uses the target software application.

9. An electronic device, characterized in that, including: A memory, configured to store a computer program; A processor connected to the memory, configured to execute the computer program stored in the memory to implement the theme skin recommendation method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, A computer program is stored in the computer-readable storage medium, and when the computer program runs, it executes the theme skin recommendation method according to any one of claims 1-7.

Images