Information generation method and device, equipment and medium

Abstract

The embodiment of the invention discloses an information generation method and device, equipment and a medium. A specific embodiment of the method comprises the steps of inputting business detailed page real-time click data and a corresponding user historical click data sequence into a word embedding layer to obtain business detailed page real-time click word embedding information and a user historical click word embedding information sequence; inputting the user historical click word embedding information sequence into a high-order click behavior feature extraction layer to generate a user high-order click behavior feature information sequence; inputting the user high-order click behavior feature information sequence into a user historical click overall feature extraction layer to generate user historical click overall feature information; inputting the user history click overall feature information and the business detail page real-time click word embedding information into a feature information cross processing layer to generate feature cross information; and generating user intention identification information. The embodiment is related to artificial intelligence, and the user intention identification information can be accurately generated, so that accurate recommended article information can be conveniently generated.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of computer technology, and in particular to information generation methods, devices, equipment, and media. Background Art

[0002] At present, with the continuous development of artificial intelligence, the refined operation of users based on artificial intelligence has become the main research and development direction of various industries. For the generation of recommended item information, the usual method is: first, determine the content that the user to be recommended likes. Then, based on the content of interest, recall the user's favorite item information set from the recall source.

[0003] However, the inventors have found that when the above method is used to generate recommended item information, the following technical problems often occur:

[0004] The item information required by the current user to be recommended cannot be accurately and effectively determined based only on the preferences of the user to be recommended.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the inventive concept and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the invention

[0006] The content of this disclosure is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this disclosure is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.

[0007] Some embodiments of the present disclosure propose information generation methods, devices, equipment and media to solve the technical problems mentioned in the above background technology section.

[0008] In a first aspect, some embodiments of the present disclosure provide an information generation method, comprising: inputting pre-acquired real-time click data of a business details page and a corresponding sequence of user historical click data into a word embedding layer to obtain real-time click word embedding information of the business details page and a sequence of user historical click word embedding information, wherein a user intention recognition model comprises: the above-mentioned word embedding layer, a high-order click behavior feature extraction layer, a user historical click overall feature extraction layer and a feature information cross-processing layer; inputting the above-mentioned user historical click word embedding information sequence into the above-mentioned high-order click behavior feature extraction layer to generate a user high-order click behavior feature information sequence; inputting the above-mentioned user high-order click behavior feature information sequence into the above-mentioned user historical click overall feature extraction layer to generate user historical click overall feature information; inputting the above-mentioned user historical click overall feature information and the above-mentioned business details page real-time click word embedding information into the above-mentioned feature information cross-processing layer to generate feature cross-information; and generating user intention recognition information based on the above-mentioned feature cross-information.

[0009] Optionally, the user intention recognition model also includes: an intention recognition information output layer; and generating user intention recognition information based on the feature cross-information, including: obtaining a user attribute information set for the real-time click data of the business details page; inputting each user attribute information in the user attribute information set into the word embedding layer to generate user attribute feature information and obtain a user attribute feature information set; performing information fusion processing on the user attribute feature information set and the feature cross-information to obtain fused feature information; inputting the fused feature information into the intention recognition information output layer to output the user intention recognition information.

[0010] Optionally, the method further includes: generating an item recall source weight information set for the user intention identification information; generating a recalled item information set based on the item recall source weight information set; sorting the item information in the recalled item information set to generate a recalled item information sequence; and recommending at least one recalled item information in the recalled item information sequence to the corresponding user terminal.

[0011] Optionally, the above-mentioned user intention identification information includes: item same-product intention score information, item same type intention score information, scene intention score information; and the above-mentioned item recall source weight information set generated for the above-mentioned user intention identification information includes: generating item recall source weight information for the above-mentioned item same-product intention score information; generating item recall source weight information for the above-mentioned item same type intention score information; generating item recall source weight information for the above-mentioned scene intention score information.

[0012] Optionally, the above-mentioned high-order click behavior feature extraction layer is a self-attention mechanism layer; and the above-mentioned inputting the above-mentioned user historical click word embedding information sequence into the above-mentioned high-order click behavior feature extraction layer to generate the above-mentioned user high-order click behavior feature information sequence, including: inputting the above-mentioned user historical click word embedding information sequence into the above-mentioned self-attention mechanism layer to generate the above-mentioned user high-order click behavior feature information sequence.

[0013] Optionally, the above-mentioned user historical click overall feature extraction layer is an attention mechanism layer based on time decay; and the above-mentioned user high-order click behavior feature information sequence is input into the above-mentioned user historical click overall feature extraction layer to generate the user historical click overall feature information, including: inputting the above-mentioned user high-order click behavior feature information sequence into the above-mentioned attention mechanism layer based on time decay to generate the above-mentioned user historical click overall feature information.

[0014] Optionally, the above-mentioned overall feature information of the user's historical clicks and the real-time click word embedding information of the business details page are input into the above-mentioned feature information cross-processing layer to generate feature cross-information, including: utilizing the above-mentioned feature information cross-processing layer to perform feature information cross-multiplication processing on the above-mentioned overall feature information of the user's historical clicks and the real-time click word embedding information of the business details page to obtain cross-product feature information; utilizing the above-mentioned feature information cross-processing layer to perform feature information subtraction processing on the above-mentioned overall feature information of the user's historical clicks and the real-time click word embedding information of the business details page to obtain subtracted feature information; performing feature information fusion on the above-mentioned cross-product feature information and the above-mentioned subtracted feature information to generate the above-mentioned feature cross-information.

[0015] Optionally, the user intention recognition model is trained by the following steps: obtaining a real-time click data set of business details pages and a user intention information set, wherein the real-time click data of business details pages include: real-time click data of business details pages, a corresponding sequence of user historical click data and a corresponding set of user attribute information; selecting real-time click data of business details pages from the real-time click data set of business details pages as target real-time click data of business details pages, and performing the following training steps: inputting the real-time click data of business details pages and a sequence of user historical click data included in the target real-time click data of business details pages into an initial word embedding layer to obtain target business details page real-time click word embedding information and a sequence of target user historical click word embedding information; inputting the target user historical click word embedding information sequence into an initial high-order click behavior feature extraction layer to generate a target user high-order click behavior feature information sequence; inputting the target user high-order click behavior feature information sequence into an initial user historical click overall feature extraction layer to generate target user historical click overall feature information; inputting the target user historical click overall feature information and the target business details page real-time click word embedding information into an initial feature information extraction layer; The method comprises the steps of: inputting the user attribute vector set included in the target business detail page real-time click data into the initial word embedding layer to obtain the target user attribute feature information set; performing information fusion processing on the target feature cross information and the target user attribute feature information set to obtain the target fusion feature information; inputting the target fusion feature information into the initial intention recognition information output layer to output the target user intention recognition information; generating the intention loss information according to the user intention information corresponding to the target business detail page real-time click data and the target user intention recognition information; determining whether the initial user intention recognition model has been trained according to the intention loss information; in response to the training being completed, determining the initial user intention recognition model as the user intention recognition model; in response to the training being incomplete, updating the model parameters of the initial user intention recognition model according to the intention loss information to generate an updated user intention recognition model as the initial user intention recognition model, and reselecting the target business detail page real-time click data from the business detail page real-time click data set to continue to perform the training step.

[0016] In a second aspect, some embodiments of the present disclosure provide an information generating device, comprising: a first input unit, configured to input pre-acquired real-time click data of a business detail page and a corresponding sequence of user historical click data into a word embedding layer, to obtain real-time word embedding information of the business detail page and a sequence of user historical click word embedding information, wherein a user intention recognition model comprises: the above-mentioned word embedding layer, a high-order click behavior feature extraction layer, a user historical click overall feature extraction layer and a feature information cross-processing layer; a second input unit, configured to input the above-mentioned user historical click word embedding information sequence into the above-mentioned high-order click behavior feature extraction layer, to generate a user high-order click behavior feature information sequence; a third input unit, configured to input the above-mentioned user high-order click behavior feature information sequence into the above-mentioned user historical click overall feature extraction layer, to generate user historical click overall feature information; a fourth input unit, configured to input the above-mentioned user historical click overall feature information and the above-mentioned business detail page real-time click word embedding information into the above-mentioned feature information cross-processing layer, to generate feature cross-information; a generating unit, configured to generate user intention recognition information according to the above-mentioned feature cross-information.

[0017] Optionally, the user intention recognition model also includes: an intention recognition information output layer; and the generation unit can be configured to: obtain a user attribute information set for the real-time click data of the business details page; input each user attribute information in the user attribute information set into the word embedding layer to generate user attribute feature information and obtain a user attribute feature information set; perform information fusion processing on the user attribute feature information set and the feature cross information to obtain fused feature information; input the fused feature information into the intention recognition information output layer to output the user intention recognition information.

[0018] Optionally, the device also includes: generating an item recall source weight information set for the above-mentioned user intention identification information; generating a recalled item information set based on the above-mentioned item recall source weight information set; sorting the item information in the above-mentioned recalled item information set to generate a recalled item information sequence; and recommending at least one recalled item information in the above-mentioned recalled item information sequence to the corresponding user terminal.

[0019] Optionally, the above-mentioned user intention identification information includes: item same-product intention score information, item same type intention score information, scene intention score information; and the generation unit can be configured to: generate item recall source weight information for the above-mentioned item same-product intention score information; generate item recall source weight information for the above-mentioned item same type intention score information; generate item recall source weight information for the above-mentioned scene intention score information.

[0020] Optionally, the above-mentioned high-order click behavior feature extraction layer is a self-attention mechanism layer; and the second input unit can be configured to: input the above-mentioned user historical click word embedding information sequence into the above-mentioned self-attention mechanism layer to generate the above-mentioned user high-order click behavior feature information sequence.

[0021] Optionally, the above-mentioned user historical click overall feature extraction layer is an attention mechanism layer based on time decay; and the third input unit can be configured to: input the above-mentioned user high-order click behavior feature information sequence into the above-mentioned attention mechanism layer based on time decay to generate the above-mentioned user historical click overall feature information.

[0022] Optionally, the fourth input unit can be configured to: utilize the above-mentioned feature information cross-processing layer to perform feature information cross-multiplication processing on the above-mentioned user history click overall feature information and the above-mentioned business details page real-time click word embedding information to obtain cross-product feature information; utilize the above-mentioned feature information cross-processing layer to perform feature information subtraction processing on the above-mentioned user history click overall feature information and the above-mentioned business details page real-time click word embedding information to obtain subtracted feature information; perform feature information fusion on the above-mentioned cross-product feature information and the above-mentioned subtracted feature information to generate the above-mentioned feature cross information.

[0023] Optionally, the user intention recognition model is trained by the following steps: obtaining a real-time click data set of business details pages and a user intention information set, wherein the real-time click data of business details pages include: real-time click data of business details pages, a corresponding sequence of user historical click data and a corresponding set of user attribute information; selecting real-time click data of business details pages from the real-time click data set of business details pages as target real-time click data of business details pages, and performing the following training steps: inputting the real-time click data of business details pages and a sequence of user historical click data included in the target real-time click data of business details pages into an initial word embedding layer to obtain target business details page real-time click word embedding information and a sequence of target user historical click word embedding information; inputting the target user historical click word embedding information sequence into an initial high-order click behavior feature extraction layer to generate a target user high-order click behavior feature information sequence; inputting the target user high-order click behavior feature information sequence into an initial user historical click overall feature extraction layer to generate target user historical click overall feature information; inputting the target user historical click overall feature information and the target business details page real-time click word embedding information into an initial feature information extraction layer; The method comprises the steps of: inputting the user attribute vector set included in the target business detail page real-time click data into the initial word embedding layer to obtain the target user attribute feature information set; performing information fusion processing on the target feature cross information and the target user attribute feature information set to obtain the target fusion feature information; inputting the target fusion feature information into the initial intention recognition information output layer to output the target user intention recognition information; generating the intention loss information according to the user intention information corresponding to the target business detail page real-time click data and the target user intention recognition information; determining whether the initial user intention recognition model has been trained according to the intention loss information; in response to the training being completed, determining the initial user intention recognition model as the user intention recognition model; in response to the training being incomplete, updating the model parameters of the initial user intention recognition model according to the intention loss information to generate an updated user intention recognition model as the initial user intention recognition model, and reselecting the target business detail page real-time click data from the business detail page real-time click data set to continue to perform the training step.

[0024] In a third aspect, some embodiments of the present disclosure provide an electronic device comprising: one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation manner in the first aspect.

[0025] In a fourth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method described in any implementation manner in the first aspect is implemented.

[0026] In a fifth aspect, some embodiments of the present disclosure provide a computer program product, including a computer program, which implements the method described in any implementation manner in the above-mentioned first aspect when executed by a processor.

[0027] The above-mentioned embodiments of the present disclosure have the following beneficial effects: the information generation method of some embodiments of the present disclosure can accurately generate user intention recognition information, so as to facilitate the subsequent generation of more accurate recommended item information. Specifically, the reason why the relevant recommended item information is not accurate enough is that the item information required by the current user to be recommended cannot be accurately and effectively determined based only on the preferred content of the user to be recommended. Based on this, the information generation method of some embodiments of the present disclosure, first, inputs the pre-acquired real-time click data of the business details page and the corresponding user history click data sequence into the word embedding layer, and obtains the real-time click word embedding information of the business details page and the user history click word embedding information sequence, wherein the user intention recognition model includes: the above-mentioned word embedding layer, the high-order click behavior feature extraction layer, the user history click overall feature extraction layer and the feature information cross-processing layer. Here, the word embedding layer can be used to convert the real-time click data of the business details page and the corresponding user history click data sequence into a word embedding form, so as to facilitate subsequent feature extraction. Then, the above-mentioned user history click word embedding information sequence is input into the above-mentioned high-order click behavior feature extraction layer to generate the user high-order click behavior feature information sequence. Here, through the high-order click behavior feature extraction layer, the high-order click feature information for the user's historical click word embedding information sequence can be accurately obtained, so that the subsequent user intention recognition information generation is more accurate. Then, the above-mentioned user high-order click behavior feature information sequence is input into the above-mentioned user historical click overall feature extraction layer to generate the user historical click overall feature information. Here, through the user historical click overall feature extraction layer, the overall feature information for the user's high-order click behavior feature information sequence (i.e., the user historical click overall feature information) can be accurately generated, and the important buried point feature information that affects the user's intention recognition can be obtained, so that the subsequent user intention recognition information generation is more accurate. Furthermore, the above-mentioned user historical click overall feature information and the above-mentioned commercial page real-time click word embedding information are input into the above-mentioned feature information cross-processing layer to generate feature cross-information. Here, through the feature information cross-processing layer, multi-faceted feature information for the user's historical click overall feature information and the above-mentioned commercial page real-time click word embedding information can be generated, and the feature cross-information obtained includes richer feature content. Finally, based on the above-mentioned feature cross-information, user intention recognition information can be accurately generated. In summary, through the user intention recognition model including the word embedding layer, the high-order click behavior feature extraction layer, the user history click overall feature extraction layer and the feature information cross-processing layer, the user intention recognition information can be accurately generated, so that the item information set required by the current user to be recommended can be generated based on the user intention recognition information. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.

[0029] Figure 1 is a schematic diagram of an application scenario of the information generation method according to some embodiments of the present disclosure;

[0030] Figure 2 are flow charts of some embodiments of the information generation method according to the present disclosure;

[0031] Figure 3 are flow charts of other embodiments of the information generating method according to the present disclosure;

[0032] Figure 4 is a schematic diagram of the structure of some embodiments of the information generating device according to the present disclosure;

[0033] Figure 5 It is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0034] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0035] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure can be combined with each other.

[0036] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0037] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0038] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0039] With regard to the collection, storage, and use of user personal information (such as real-time click data on the business detail page and user historical click data series) involved in this disclosure, relevant organizations or individuals shall, before performing corresponding operations, fulfill obligations including conducting personal information security impact assessments, fulfilling the obligation to inform the personal information subject, and obtaining the authorization and consent of the personal information subject in advance.

[0040] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0041] Figure 1 It is a schematic diagram of an application scenario of the information generation method according to some embodiments of the present disclosure.

[0042] exist Figure 1 In the application scenario, first, the electronic device 101 can input the pre-acquired real-time click data 102 of the business details page and the corresponding user history click data sequence 103 into the word embedding layer 105 to obtain the real-time click word embedding information 106 of the business details page and the user history click word embedding information sequence 107. Among them, the user intention recognition model 104 includes: the above-mentioned word embedding layer 105, the high-order click behavior feature extraction layer 108, the user history click overall feature extraction layer 110 and the feature information cross-processing layer 112. Then, the electronic device 101 can input the above-mentioned user history click word embedding information sequence 107 into the above-mentioned high-order click behavior feature extraction layer 108 to generate the user high-order click behavior feature information sequence 109. Next, the electronic device 101 can input the above-mentioned user high-order click behavior feature information sequence 109 into the above-mentioned user history click overall feature extraction layer 110 to generate the user history click overall feature information 111. Furthermore, the electronic device 101 may input the above-mentioned user history click overall feature information 111 and the above-mentioned business detail page real-time click word embedding information 106 into the above-mentioned feature information cross processing layer 112 to generate feature cross information 113. Finally, the electronic device 101 may generate user intention recognition information 114 according to the above-mentioned feature cross information 113.

[0043] It should be noted that the electronic device 101 can be hardware or software. When the electronic device is hardware, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or it can be implemented as a single server or a single terminal device. When the electronic device is embodied as software, it can be installed in the hardware devices listed above. It can be implemented as multiple software or software modules for providing distributed services, for example, or it can be implemented as a single software or software module. No specific limitation is made here.

[0044] It should be understood that Figure 1 The number of electronic devices in the embodiment is only for illustration. Any number of electronic devices may be provided according to implementation requirements.

[0045] Continue to refer Figure 2 , shows a process 200 of some embodiments of the information generation method according to the present disclosure. The information generation method comprises the following steps:

[0046] Step 201: input the pre-acquired real-time click data of the business details page and the corresponding user history click data sequence into the word embedding layer to obtain the real-time click word embedding information of the business details page and the user history click word embedding information sequence.

[0047] In some embodiments, the execution subject of the above information generation method (for example Figure 1The electronic device 101 shown can input the pre-acquired real-time click data of the business details page and the corresponding user history click data sequence into the word embedding layer to obtain the real-time click word embedding information of the business details page and the user history click word embedding information sequence. Among them, the real-time click data of the business details page can be the real-time click data of the corresponding user when clicking the target business details page. Specifically, the real-time click data may include: real-time buried point identification, real-time commodity identification, real-time category identification, and real-time search word identification. The target business details page may be the business details page of the target commodity. The user history click data sequence may be the historical click data sequence of the corresponding user in the target time period before clicking the target business details page. In practice, the user history click data sequence may be the click data set of the user within 24 hours before clicking the target business details page. The target business details page may be the business details page of the target commodity. Specifically, the click data set may include: buried point identification of the user's click behavior, commodity identification, category identification, search word identification, and click time feature data. The user intention recognition model may be a model for generating user intention recognition information. Specifically, the user intention recognition information may be the intention information of the user clicking the business details page. For example, the user intention recognition information may be the intention information of purchasing the corresponding item on the product detail page. The word embedding layer may be a network layer that performs word embedding processing on the click data. Specifically, the word embedding layer may be an Embedding layer. The real-time click word embedding information of the product detail page may be the original vector representation of the real-time click data of the product detail page. The user historical click word embedding information in the user historical click word embedding information sequence and the user historical click data in the user historical click data sequence have a one-to-one correspondence. The real-time click data of the product detail page and the corresponding click product detail page of the corresponding user historical click data sequence are the same. The user historical click word embedding information in the user historical click word embedding information sequence may be the original vector representation of the user historical click data. The high-order click behavior feature extraction layer may be a network layer that generates high-order click behavior feature information. The high-order click behavior feature information may be a higher-order vector representation of the click behavior corresponding to the user historical click data. The user high-order click behavior feature information in the user high-order click behavior feature information sequence and the user historical click word embedding information in the user historical click word embedding information sequence have a one-to-one correspondence. In practice, the high-order click behavior feature extraction layer may be a multi-layer serially connected recurrent neural network layer. The overall feature extraction layer of user history clicks can be a network layer that generates overall feature information of user history clicks. The overall feature information of user history clicks can represent the overall vector representation of the user click behavior corresponding to the user history click data sequence. In practice, the overall feature extraction layer of user history clicks can be a multi-layer serially connected recurrent neural network layer. Among them, the feature information cross-processing layer can be a network layer that performs information cross-processing on feature information. Feature cross-information can include: cross-feature information between the overall feature information of user history clicks and the above-mentioned real-time click word embedding information of the business details page.For example, the feature information cross-processing layer can be a multi-layer serially connected convolutional neural network layer.

[0048] It should be noted that the user's historical click data sequence may be a data sequence after data normalization processing according to the time window corresponding to the target time period. The closer the corresponding time of the user's historical click data is to the time of clicking the product details page, the closer the normalization result is to 0, and vice versa.

[0049] Step 202: input the above-mentioned user historical click word embedding information sequence into the above-mentioned high-order click behavior feature extraction layer to generate the user high-order click behavior feature information sequence.

[0050] In some embodiments, the execution subject may input the user's historical click word embedding information sequence into the high-order click behavior feature extraction layer to generate a user's high-order click behavior feature information sequence.

[0051] In some optional implementations of some embodiments, the above-mentioned high-order click behavior feature extraction layer is a self-attention mechanism layer (Self-attention layer).

[0052] In practice, the self-attention mechanism layer can be formulated as follows:

[0053]

[0054] Among them, z i It can be the user's high-level click behavior feature information corresponding to the i-th user's historical click data. n It can represent the user's historical click word embedding information corresponding to the nth user's historical click data. ij It can be the correlation score between the click data of the i-th user and the historical click data of the j-th user. The correlation score can be a value between 0 and 1. j It can be the user's high-level click behavior feature information corresponding to the j-th user's historical click data. V The weight matrix can be.

[0055] a ij It can be generated by the following formula:

[0056]

[0057] Among them, e ij It can be m n and m j The correlation information between them. ik Represents all e corresponding to the user's historical click data sequence ij .

[0058] eij It can be generated by the following formula:

[0059]

[0060] Among them, W Q Can be a weight matrix. W K It can be a weight matrix. d can be a feature dimension of the user's high-order click behavior feature information.

[0061] Optionally, the execution entity may input the user's historical click feature information sequence into the self-attention mechanism layer to generate the user's high-order click behavior feature information sequence.

[0062] Step 203: input the user high-order click behavior feature information sequence into the user history click overall feature extraction layer to generate user history click overall feature information.

[0063] In some embodiments, the execution entity may input the user high-order click behavior feature information sequence into the user history click overall feature extraction layer to generate user history click overall feature information.

[0064] In some optional implementations of some embodiments, the above-mentioned user history click overall feature extraction layer is an attention mechanism layer based on time decay (attention unit with time decay).

[0065] Specifically, the time-decayed attention mechanism layer can be formulated as follows:

[0066]

[0067] Among them, h can be the overall characteristic information of the user's historical clicks. a It can be a task vector that represents the current intent recognition task. j It can be the user's high-level click behavior feature information corresponding to the j-th user's historical click data. j , v a ) can be Z j The corresponding attention weights.

[0068] a(z j , v a ) is generated by the following formula:

[0069]

[0070] Among them, CONCAT(v a , z j ) can be v a and Z jPerform vector concatenation. MLP(CONCAT(v a , z j ) can be to input the vector concatenation result into the fully connected layer to obtain the fully connected result. j It can be the normalized time information corresponding to the historical click data of the jth user. λ can be a hyperparameter.

[0071] Optionally, the execution entity may input the user's high-order click behavior feature information sequence into the time-decay-based attention mechanism layer to generate the overall feature information of the user's historical clicks.

[0072] Step 204: input the overall feature information of the user's historical clicks and the real-time click word embedding information of the business detail page into the feature information cross-processing layer to generate feature cross-information.

[0073] In some embodiments, the execution entity may input the overall feature information of the user's historical clicks and the real-time click word embedding information of the business details page into the feature information cross-processing layer to generate feature cross-information.

[0074] In some optional implementations of some embodiments, the inputting of the overall feature information of the user's historical clicks and the real-time click word embedding information of the business detail page into the feature information cross-processing layer to generate feature cross-information may include the following steps:

[0075] In the first step, the feature information cross-processing layer is used to perform feature information cross-product processing on the overall feature information of the user's historical clicks and the real-time click word embedding information of the business details page to obtain cross-product feature information.

[0076] As an example, the execution entity may utilize the feature information cross-processing layer to perform vector cross-product processing on the vector corresponding to the overall feature information of the user's historical clicks and the vector corresponding to the real-time click word embedding information of the business details page to generate a cross-product vector as the cross-product feature information.

[0077] In the second step, the feature information cross-processing layer is used to perform feature information subtraction processing on the overall feature information of the user's historical clicks and the feature information of the real-time clicks on the business details page to obtain subtracted feature information.

[0078] As an example, the execution entity may utilize the feature information cross-processing layer to perform vector subtraction processing on the corresponding vector of the overall feature information of the user's historical clicks and the corresponding vector of the real-time click word embedding information of the business details page to generate a subtraction vector as the subtraction feature information.

[0079] The third step is to fuse the cross product feature information and the subtraction feature information to generate the feature cross information.

[0080] As an example, the execution subject may perform vector concatenation of a vector corresponding to the cross product feature information and a vector corresponding to the subtraction feature information to obtain a concatenated vector as feature cross information.

[0081] In some optional implementations of some embodiments, the user intent recognition model is trained by the following steps:

[0082] The first step is to obtain a real-time click data set of the business details page and a user intention information set. The real-time click data of the business details page includes: real-time click data of the business details page, a corresponding user historical click data sequence, and a corresponding user attribute information set. There is a one-to-one correspondence between the real-time click data of the business details page in the real-time click data set of the business details page and the user intention data in the user intention information set.

[0083] In the second step, the real-time click data of the business details page is selected from the above-mentioned real-time click data set of the business details page as the real-time click data of the target business details page, and the following training steps are performed:

[0084] Sub-step 1: Input the real-time click data of the business details page and the user history click data sequence included in the above target business details page real-time click data into the initial word embedding layer to obtain the target business details page real-time click word embedding information and the target user history click word embedding information sequence. The initial user intent recognition model can be a user intent recognition model that has not yet been trained. The initial word embedding layer can be a word embedding layer that has not yet been trained.

[0085] Sub-step 2: Input the above target user historical click word embedding information sequence into the initial high-order click behavior feature extraction layer to generate the target user high-order click behavior feature information sequence. The initial high-order click behavior feature extraction layer can be a high-order click behavior feature extraction layer that has not yet been trained.

[0086] Sub-step 3: input the target user's high-order click behavior feature information sequence into the initial user history click overall feature extraction layer to generate the target user history click overall feature information. The initial user history click overall feature extraction layer can be a user history click overall feature extraction layer that has not yet been trained.

[0087] Sub-step 4: Input the above target user historical click overall feature information and the above target business detail page real-time click word embedding information into the initial feature information cross-processing layer to generate target feature cross-processing information. The initial feature information cross-processing layer can be a feature information cross-processing layer for which the network has not yet been trained.

[0088] Sub-step 5: input the user attribute vector set included in the above-mentioned real-time click data of the target business details page into the above-mentioned initial word embedding layer to obtain the target user attribute vector set.

[0089] Sub-step 6: fusing the target feature cross information and the target user attribute vector set to obtain target fused feature information.

[0090] Sub-step 7, input the above-mentioned target fusion feature information into the initial intention recognition information output layer to output the above-mentioned target user intention recognition information. Among them, the initial intention recognition information output layer can be the intention recognition information output layer of the network that has not yet been trained. In practice, the intention recognition information output layer can be a network layer that outputs the intention recognition information. In practice, the intention recognition information output layer can be a fully connected layer with multiple layers connected in series.

[0091] Sub-step 8: Generate intention loss information based on the user intention information corresponding to the real-time click data of the target business details page and the target user intention recognition information. The intention recognition information may represent the difference between the user intention information corresponding to the real-time click data of the target business details page and the target user intention recognition information.

[0092] As an example, the execution entity may use a cross entropy loss function to determine intent loss information for the user intent information and the target user intent recognition information.

[0093] Sub-step 9, determining whether the initial user intent recognition model has been trained based on the intent loss information.

[0094] As an example, in response to determining that the intention loss information is less than or equal to a preset value, it is determined that the initial user intention recognition model training is completed. In response to determining that the intention loss information is greater than a preset value, it is determined that the initial user intention recognition model training is not completed.

[0095] Sub-step 10, in response to the training being completed, determining the above initial user intention recognition model as the user intention recognition model.

[0096] In the second step, in response to the incomplete training, the model parameters of the initial user intent recognition model are updated according to the intention loss information to generate an updated user intent recognition model as the initial user intent recognition model, and the target product details page real-time click data is reselected from the product details page real-time click data set to continue the training step.

[0097] Step 205: Generate user intention recognition information based on the above feature cross information.

[0098] In some embodiments, the execution entity may generate user intention recognition information based on the feature cross information, wherein the user intention recognition information may be the user intention recognition information of clicking on the business details page.

[0099] In some optional implementations of some embodiments, after step 205, the steps further include:

[0100] The first step is to generate an item recall source weight information set for the above-mentioned user intention identification information. Among them, the weight information of the item recall source in the item recall source weight information set. The weight information can be a numerical value between 0 and 1. The item recall source can be the source channel of the recalled item. For example, the item recall source can be a target recall channel through which the item information is recalled. For example, the target recall channel can be a mobile phone channel. The item recall source weight information in the item recall source weight information set can have a one-to-one correspondence with the item recall source in the item recall source set. The item recall source weight information can characterize the recall importance of the corresponding item recall source.

[0101] In the second step, a recalled item information set is generated according to the item recall source weight information set. The recalled item information in the recalled item information set may be item information of the recalled item. The recalled item may be an item that has been recalled. The item information may be item noun information of the recalled item.

[0102] As an example, first, the execution subject may set the item recall quantity of the corresponding item recall source according to the recall source weight information corresponding to each item recall source. Then, for each item recall source, the item information corresponding to the item recall quantity is recalled as the recalled item information.

[0103] The third step is to sort the item information of the recalled item information set to generate a recalled item information sequence. The individual recalled item information in the recalled item information sequence may be a subset of the information in the recalled item information set. Similarly, the individual recalled item information in the recalled item information sequence may be equivalent to the individual recalled item information included in the recalled item information set.

[0104] As an example, the execution subject may use an item information sorting model to sort the item information set to generate a recalled item information sequence. The item information sorting model may be a neural network model for sorting item information. For example, the item information sorting model may be a Transformer model.

[0105] The fourth step is to recommend at least one recalled item information in the recalled item information sequence to a corresponding user terminal.

[0106] As an example, the execution entity may recommend the recalled item information located at the first target number of items in the recalled item information sequence as at least one recalled item information to the corresponding user terminal.

[0107] In some optional implementations of some embodiments, the above-mentioned user intention identification information includes: item same-product intention score information, item same-product intention score information, and scene intention score information. Among them, the item same-product intention score information can characterize the importance of the recall source corresponding to the item same-product recall source. The item same-product recall source can be a recall of items of the same product as the target item. In practice, items of the same product as the target item can be items of the same type as the target item, or items that are similar to the target item. The item same-product intention score information can characterize the importance of the recall source corresponding to the item same-product recall source. The item same-product recall source can be a recall of items of the same category as the target item. The scene intention score information can characterize the importance of the recall source corresponding to the item scene recall source. The item scene recall source can be an item in the same scene as the target item.

[0108] Optionally, the generating of the item recall source weight information set for the user intention identification information may include the following steps:

[0109] The first step is to generate item recall source weight information for the above item same-product intention score information.

[0110] As an example, the execution entity may normalize the item same-product intention score information to generate item recall source weight information.

[0111] The second step is to generate item recall source weight information for the same type of intention score information of the above items.

[0112] As an example, the execution entity may normalize the item same intention score information to generate item recall source weight information.

[0113] The third step is to generate item recall source weight information for the above scenario intention score information.

[0114] As an example, the execution entity may normalize the scene intention score information to generate item recall source weight information.

[0115] The above-mentioned embodiments of the present disclosure have the following beneficial effects: the information generation method of some embodiments of the present disclosure can accurately generate user intention recognition information, so as to facilitate the subsequent generation of more accurate recommended item information. Specifically, the reason why the relevant recommended item information is not accurate enough is that the item information required by the current user to be recommended cannot be accurately and effectively determined based only on the preferred content of the user to be recommended. Based on this, the information generation method of some embodiments of the present disclosure, first, inputs the pre-acquired real-time click data of the business details page and the corresponding user history click data sequence into the word embedding layer, and obtains the real-time click word embedding information of the business details page and the user history click word embedding information sequence, wherein the user intention recognition model includes: the above-mentioned word embedding layer, the high-order click behavior feature extraction layer, the user history click overall feature extraction layer and the feature information cross-processing layer. Here, the word embedding layer can be used to convert the real-time click data of the business details page and the corresponding user history click data sequence into a word embedding form, so as to facilitate subsequent feature extraction. Then, the above-mentioned user history click word embedding information sequence is input into the above-mentioned high-order click behavior feature extraction layer to generate the user high-order click behavior feature information sequence. Here, through the high-order click behavior feature extraction layer, the high-order click feature information for the user's historical click word embedding information sequence can be accurately obtained, so that the subsequent user intention recognition information generation is more accurate. Then, the above-mentioned user high-order click behavior feature information sequence is input into the above-mentioned user historical click overall feature extraction layer to generate the user historical click overall feature information. Here, through the user historical click overall feature extraction layer, the overall feature information for the user's high-order click behavior feature information sequence (i.e., the user historical click overall feature information) can be accurately generated, and the important buried point feature information that affects the user's intention recognition can be obtained, so that the subsequent user intention recognition information generation is more accurate. Furthermore, the above-mentioned user historical click overall feature information and the above-mentioned commercial page real-time click word embedding information are input into the above-mentioned feature information cross-processing layer to generate feature cross-information. Here, through the feature information cross-processing layer, multi-faceted feature information for the user's historical click overall feature information and the above-mentioned commercial page real-time click word embedding information can be generated, and the feature cross-information obtained includes richer feature content. Finally, based on the above-mentioned feature cross-information, user intention recognition information can be accurately generated. In summary, through the user intention recognition model including the word embedding layer, the high-order click behavior feature extraction layer, the user history click overall feature extraction layer and the feature information cross-processing layer, the user intention recognition information can be accurately generated, so that the item information set required by the current user to be recommended can be generated based on the user intention recognition information.

[0116] Further references Figure 3 , shows a process 300 of another embodiment of the information generation method according to the present disclosure. The information generation method comprises the following steps:

[0117] Step 301, the pre-acquired real-time click data of the business details page and the corresponding user history click data sequence are input into the word embedding layer to obtain the real-time click word embedding information of the business details page and the user history click word embedding information sequence.

[0118] Step 302: input the above-mentioned user historical click word embedding information sequence into the above-mentioned high-order click behavior feature extraction layer to generate the user high-order click behavior feature information sequence.

[0119] Step 303: input the user high-order click behavior feature information sequence into the user history click overall feature extraction layer to generate user history click overall feature information.

[0120] Step 304: input the overall feature information of the user's historical clicks and the real-time click word embedding information of the business detail page into the feature information cross-processing layer to generate feature cross-information.

[0121] In some embodiments, the specific implementation of steps 301-304 and the technical effects thereof can be referred to in Figure 2 The steps 201-204 in the corresponding embodiment are not described in detail here.

[0122] Step 305: Obtain a user attribute information set for the real-time click data of the business detail page.

[0123] In some embodiments, an execution entity (e.g. Figure 1 The electronic device 101 shown can obtain a user attribute information set for the real-time click data of the business detail page in a wired or wireless manner. The user attribute information can be an attribute value corresponding to a user attribute. For example, the user attribute can be, but is not limited to, one of the following: user gender, user age, user job. The user intent recognition model includes an intent recognition information output layer.

[0124] Step 306: input each user attribute information in the above user attribute information set into the above word embedding layer to generate user attribute feature information and obtain a user attribute feature information set.

[0125] In some embodiments, the execution subject may input each user attribute information in the user attribute information set into the word embedding layer to generate user attribute feature information and obtain a user attribute feature information set. The user attribute feature information may represent feature information of user attributes. Specifically, the user attribute feature information may be in vector form.

[0126] Step 307: perform information fusion processing on the user attribute vector set and the feature cross information to obtain fused feature information.

[0127] In some embodiments, the execution entity may perform information fusion processing on the user attribute vector set and the feature cross information to obtain fused feature information.

[0128] Step 308: input the fused feature information into the intention recognition information output layer to output the user intention recognition information.

[0129] In some embodiments, the execution entity may input the fused feature information into the intention recognition information output layer to output the user intention recognition information.

[0130] from Figure 3 It can be seen that Figure 2 Compared with the description of some corresponding embodiments, Figure 3 Correspondingly, process 300 of the information generation method in some embodiments can enrich the feature information input into the user intention recognition model by considering the user attribute feature information set corresponding to the user attribute information set, so that user intention recognition information can be generated more accurately using the user intention recognition model.

[0131] Further references Figure 4 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of an information generating device, and these device embodiments are Figure 2 Corresponding to the method embodiments shown, the information generating device can be specifically applied to various electronic devices.

[0132] like Figure 4As shown, an information generating device 400 includes: a first input unit 401, a second input unit 402, a third input unit 403, a fourth input unit 404 and a generating unit 405. Among them, the first input unit 401 is configured to input the pre-acquired real-time click data of the business details page and the corresponding user history click data sequence into the word embedding layer to obtain the real-time click word embedding information of the business details page and the user history click word embedding information sequence, wherein the user intention recognition model includes: the above-mentioned word embedding layer, the high-order click behavior feature extraction layer, the user history click overall feature extraction layer and the feature information cross-processing layer; the second input unit 402 is configured to input the above-mentioned user history click word embedding information sequence into the above-mentioned high-order click behavior feature extraction layer to generate the user high-order click behavior feature information sequence; the third input unit 403 is configured to input the above-mentioned user high-order click behavior feature information sequence into the above-mentioned user history click overall feature extraction layer to generate the user history click overall feature information; the fourth input unit 404 is configured to input the above-mentioned user history click overall feature information and the above-mentioned business details page real-time click word embedding information into the above-mentioned feature information cross-processing layer to generate feature cross information; the generation unit 405 is configured to generate user intention recognition information according to the above-mentioned feature cross information.

[0133] In some optional implementations of some embodiments, the user intent recognition model also includes: an intent recognition information output layer; and the generation unit 405 can be further configured to: obtain a user attribute information set for the real-time click data of the above-mentioned business details page; input each user attribute information in the above-mentioned user attribute information set into the above-mentioned word embedding layer to generate user attribute feature information and obtain a user attribute feature information set; perform information fusion processing on the above-mentioned user attribute feature information set and the above-mentioned feature cross information to obtain fused feature information; input the above-mentioned fused feature information into the above-mentioned intent recognition information output layer to output the above-mentioned user intent recognition information.

[0134] In some optional implementations of some embodiments, the generation unit 405 may be further configured to: generate an item recall source weight information set for the above-mentioned user intention identification information; generate a recalled item information set based on the above-mentioned item recall source weight information set; sort the item information in the above-mentioned recalled item information set to generate a recalled item information sequence; and recommend at least one recalled item information in the above-mentioned recalled item information sequence to the corresponding user terminal.

[0135] In some optional implementations of some embodiments, the above-mentioned user intention identification information includes: item same-product intention score information, item same type intention score information, scene intention score information; and the generation unit 405 can be further configured to: generate item recall source weight information for the above-mentioned item same-product intention score information; generate item recall source weight information for the above-mentioned item same type intention score information; generate item recall source weight information for the above-mentioned scene intention score information.

[0136] In some optional implementations of some embodiments, the above-mentioned high-order click behavior feature extraction layer is a self-attention mechanism layer; and the second input unit 402 can be further configured to: input the above-mentioned user historical click word embedding information sequence into the above-mentioned self-attention mechanism layer to generate the above-mentioned user high-order click behavior feature information sequence.

[0137] In some optional implementations of some embodiments, the above-mentioned user historical click overall feature extraction layer is an attention mechanism layer based on time decay; and the third input unit 403 can be further configured to: input the above-mentioned user high-order click behavior feature information sequence into the above-mentioned attention mechanism layer based on time decay to generate the above-mentioned user historical click overall feature information.

[0138] In some optional implementations of some embodiments, the fourth input unit 404 can be further configured to: utilize the above-mentioned feature information cross-processing layer to perform feature information cross-multiplication processing on the above-mentioned user historical click overall feature information and the above-mentioned business details page real-time click word embedding information to obtain cross-product feature information; utilize the above-mentioned feature information cross-processing layer to perform feature information subtraction processing on the above-mentioned user historical click overall feature information and the above-mentioned business details page real-time click word embedding information to obtain subtracted feature information; perform feature information fusion on the above-mentioned cross-product feature information and the above-mentioned subtracted feature information to generate the above-mentioned feature cross information.

[0139] In some optional implementations of some embodiments, the user intent recognition model is trained by the following steps: obtaining a real-time click data set of business details pages and a user intent information set, wherein the real-time click data of the business details pages includes: real-time click data of the business details pages, a corresponding sequence of user historical click data, and a corresponding user attribute information set; selecting the real-time click data of the business details pages from the real-time click data set of the business details pages as the target real-time click data of the business details pages, and performing the following training steps: inputting the real-time click data of the business details pages and the sequence of user historical click data included in the target real-time click data of the business details pages into the initial word embedding layer to obtain the real-time click word embedding information of the target business details pages and the sequence of target user historical click word embedding information; inputting the target user historical click word embedding information sequence into the initial high-order click behavior feature extraction layer to generate a target user high-order click behavior feature information sequence; inputting the target user high-order click behavior feature information sequence into the initial user historical click overall feature extraction layer to generate the target user historical click overall feature information; inputting the target user historical click overall feature information and the target business details page real-time click word embedding information into the initial high-order click behavior feature extraction layer to generate the target user high-order click behavior feature information sequence; the target business details page real-time click data and the target user intention recognition information set are input into the initial feature information cross-processing layer to generate target feature cross-information; the user attribute vector set included in the target business details page real-time click data is input into the initial word embedding layer to obtain the target user attribute feature information set; the target feature cross-information and the target user attribute feature information set are subjected to information fusion processing to obtain target fusion feature information; the target fusion feature information is input into the initial intention recognition information output layer to output target user intention recognition information; according to the user intention information corresponding to the target business details page real-time click data and the target user intention recognition information, intention loss information is generated; according to the intention loss information, it is determined whether the initial user intention recognition model is trained; in response to the training being completed, the initial user intention recognition model is determined as the user intention recognition model; in response to the training being incomplete, the model parameters of the initial user intention recognition model are updated according to the intention loss information to generate an updated user intention recognition model as the initial user intention recognition model, and the target business details page real-time click data set is reselected to continue to perform the training step.

[0140] It is understandable that the units recorded in the information generating device 400 are similar to those in the reference Figure 2 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the information generating device 400 and the units included therein, and will not be described in detail here.

[0141] Reference below Figure 5 , which shows an electronic device (eg, Figure 1Schematic diagram of the structure of the electronic device 101)500. Figure 5 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0142] like Figure 5 As shown, the electronic device 500 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory 502 or a program loaded from a storage device 508 to a random access memory 503. Various programs and data required for the operation of the electronic device 500 are also stored in the random access memory 503. The processing device 501, the read-only memory 502, and the random access memory 503 are connected to each other via a bus 504. An input / output interface 505 is also connected to the bus 504.

[0143] Typically, the following devices may be connected to the input / output interface 505: an input device 506 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 507 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 508 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 509. The communication device 509 may allow the electronic device 500 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 5 The electronic device 500 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 5 Each block shown in the figure may represent one device, or may represent multiple devices as required.

[0144] In particular, according to some embodiments of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, some embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from the network through the communication device 509, or installed from the storage device 508, or installed from the read-only memory 502. When the computer program is executed by the processing device 501, the above-mentioned functions defined in the method of some embodiments of the present disclosure are executed.

[0145] It should be noted that the computer-readable medium in some embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In some embodiments of the present disclosure, the computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0146] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0147] The computer-readable medium may be included in the electronic device; or it may exist independently without being installed in the electronic device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: inputs the pre-acquired real-time click data of the business details page and the corresponding user history click data sequence into the word embedding layer, and obtains the real-time click word embedding information of the business details page and the user history click word embedding information sequence, wherein the user intention recognition model includes: the word embedding layer, the high-order click behavior feature extraction layer, the user history click overall feature extraction layer and the feature information cross-processing layer; inputs the user history click word embedding information sequence into the high-order click behavior feature extraction layer to generate the user high-order click behavior feature information sequence; inputs the user high-order click behavior feature information sequence into the user history click overall feature extraction layer to generate the user history click overall feature information; inputs the user history click overall feature information and the business details page real-time click word embedding information into the feature information cross-processing layer to generate feature cross information; generates user intention recognition information according to the feature cross information.

[0148] Computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0149] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0150] The units described in some embodiments of the present disclosure may be implemented by software or by hardware. The described units may also be provided in a processor, for example, may be described as: a processor comprising a first input unit, a second input unit, a third input unit, a fourth input unit and a generation unit. The names of these units do not, in some cases, constitute limitations on the units themselves, for example, the generation unit may also be described as "a unit for generating user intent recognition information based on the above-mentioned feature cross information".

[0151] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0152] Some embodiments of the present disclosure further provide a computer program product, including a computer program, which implements any of the above-mentioned information generation methods when executed by a processor.

[0153] The above descriptions are only some preferred embodiments of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with the technical features with similar functions disclosed in the embodiments of the present disclosure (but not limited to) and the technical solutions formed.

Claims

1. A method for generating information, comprising: Input the pre-acquired real-time click data of the business details page and the corresponding user history click data sequence into the word embedding layer to obtain the real-time click word embedding information of the business details page and the user history click word embedding information sequence, wherein the user intention recognition model includes: the word embedding layer, the high-order click behavior feature extraction layer, the user history click overall feature extraction layer and the feature information cross-processing layer; Inputting the user's historical click word embedding information sequence into the high-order click behavior feature extraction layer to generate a user's high-order click behavior feature information sequence; Inputting the user high-order click behavior feature information sequence into the user history click overall feature extraction layer to generate user history click overall feature information; Inputting the overall feature information of the user's historical clicks and the real-time click word embedding information of the business detail page into the feature information cross-processing layer to generate feature cross-information; User intention recognition information is generated based on the feature cross information.

2. The method according to claim 1, wherein: The user intention recognition model further includes: an intention recognition information output layer; and Generating user intention recognition information according to the feature cross information includes: Obtaining a user attribute information set for real-time click data on the business details page; Inputting each user attribute information in the user attribute information set into the word embedding layer to generate user attribute feature information, thereby obtaining a user attribute feature information set; Performing information fusion processing on the user attribute feature information set and the feature cross information to obtain fused feature information; The fused feature information is input into the intention recognition information output layer to output the user intention recognition information.

3. The method according to claim 1, wherein: The method further comprises: generating an item recall source weight information set for the user intention identification information; Generate a recalled item information set according to the item recall source weight information set; Sorting the recalled item information set to generate a recalled item information sequence; At least one piece of recalled item information in the recalled item information sequence is recommended to a corresponding user terminal.

4. The method according to claim 3, wherein: The user intention recognition information includes: item same product intention score information, item same category intention score information, scene intention score information; and The generating of the item recall source weight information set for the user intention identification information includes: Generate item recall source weight information for the item same product intention score information; Generate item recall source weight information for the item similar intention score information; Generate item recall source weight information for the scene intent score information.

5. The method according to claim 1, wherein: The high-order click behavior feature extraction layer is a self-attention mechanism layer; as well as The step of inputting the user's historical click word embedding information sequence into the high-order click behavior feature extraction layer to generate a user's high-order click behavior feature information sequence includes: The user's historical click word embedding information sequence is input into the self-attention mechanism layer to generate the user's high-order click behavior feature information sequence.

6. The method according to claim 1, wherein: The user history click overall feature extraction layer is an attention mechanism layer based on time decay; and The step of inputting the user high-order click behavior feature information sequence into the user history click overall feature extraction layer to generate the user history click overall feature information includes: The user's high-order click behavior feature information sequence is input into the time-decay-based attention mechanism layer to generate the overall feature information of the user's historical clicks.

7. The method according to claim 1, wherein: The step of inputting the overall feature information of the user's historical clicks and the real-time click word embedding information of the business detail page into the feature information cross-processing layer to generate feature cross-information includes: Using the feature information cross processing layer, perform feature information cross product processing on the overall feature information of the user's historical clicks and the real-time click word embedding information of the business details page to obtain cross product feature information; Using the feature information cross-processing layer, feature information subtraction processing is performed on the overall feature information of the user's historical clicks and the real-time click word embedding information of the business details page to obtain subtracted feature information; The cross product feature information and the subtraction feature information are fused to generate the feature cross information.

8. The method according to claim 1, wherein: The user intent recognition model is trained through the following steps: Acquire a real-time click data set of a business details page and a user intention information set, wherein the real-time click data of the business details page includes: real-time click data of the business details page, a corresponding user historical click data sequence and a corresponding user attribute information set; Select the real-time click data of the business details page from the real-time click data set of the business details page as the real-time click data of the target business details page, and perform the following training steps: Inputting the target business detail page real-time click data and the user history click data sequence included in the target business detail page real-time click data into the initial word embedding layer to obtain the target business detail page real-time click word embedding information and the target user history click word embedding information sequence; Inputting the target user's historical click word embedding information sequence into the initial high-order click behavior feature extraction layer to generate the target user's high-order click behavior feature information sequence; Inputting the target user's high-order click behavior feature information sequence into the initial user's historical click overall feature extraction layer to generate the target user's historical click overall feature information; Inputting the overall feature information of the target user's historical clicks and the real-time click word embedding information of the target business detail page into the initial feature information cross-processing layer to generate target feature cross-information; Inputting the user attribute vector set included in the real-time click data of the target business details page into the initial word embedding layer to obtain a target user attribute feature information set; Performing information fusion processing on the target feature cross information and the target user attribute feature information set to obtain target fusion feature information; Inputting the target fusion feature information into the initial intention recognition information output layer to output the target user intention recognition information; Generate intention loss information according to the user intention information corresponding to the real-time click data of the target business details page and the target user intention identification information; Determining whether the initial user intent recognition model has been trained according to the intent loss information; In response to the training being completed, determining the initial user intent recognition model as a user intent recognition model; In response to the training being incomplete, the model parameters of the initial user intent recognition model are updated according to the intent loss information to generate an updated user intent recognition model as the initial user intent recognition model, and the target business details page real-time click data is reselected from the business details page real-time click data set to continue executing the training step.

9. An information generating device, comprising: The first input unit is configured to input the pre-acquired real-time click data of the business details page and the corresponding user history click data sequence into the word embedding layer to obtain the real-time click word embedding information of the business details page and the user history click word embedding information sequence, wherein the user intention recognition model includes: the word embedding layer, the high-order click behavior feature extraction layer, the user history click overall feature extraction layer and the feature information cross-processing layer; A second input unit is configured to input the user's historical click word embedding information sequence into the high-order click behavior feature extraction layer to generate a user's high-order click behavior feature information sequence; A third input unit is configured to input the user high-order click behavior feature information sequence into the user history click overall feature extraction layer to generate user history click overall feature information; a fourth input unit configured to input the overall feature information of the user's historical clicks and the real-time click word embedding information of the business detail page into the feature information cross-processing layer to generate feature cross-information; The generating unit is configured to generate user intention recognition information according to the feature cross information.

10. An electronic device comprising: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 8.

11. A computer readable medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

12. A computer program product, comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 8.

Images