Material recommendation method, device, equipment and storage medium

Abstract

The present application discloses a material recommendation method, apparatus, device, and storage medium, relating to the field of recommendation technology. The method comprises: selecting at least one initial material from a material library based on historical interaction data of a target user; selecting at least one candidate material associated with each initial material from the material library based on material similarity data between associated material pairs; determining a material score for each candidate material, and selecting a target material from the candidate materials based on the material score; and recommending the target material to the target user. The technical solution of the present application improves the matching degree between the recommendation results and the target user, thereby enhancing the user experience of the target user.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of recommendation technology, and in particular to a material recommendation method, apparatus, device, and storage medium. Background Art

[0002] With the development of information technology and the internet, people have gradually moved from an era of information scarcity to an era of information overload. In this era, finding interesting information amidst a vast amount of information is extremely difficult. Therefore, recommendation systems have emerged to help users discover valuable information while also helping information producers increase their visibility.

[0003] However, when recommending materials in the prior art, the recommendation results have a low degree of matching with the user, making it difficult to achieve accurate recommendations, resulting in a poor user experience for the user. Summary of the Invention

[0004] The present application provides a material recommendation method, apparatus, device and storage medium to improve the matching degree between recommendation results and users, while improving recommendation efficiency.

[0005] In a first aspect, an embodiment of the present application provides a material recommendation method, comprising:

[0006] Select at least one initial material from the material library based on the historical interaction data of the target user;

[0007] selecting, from the material library, at least one candidate material associated with each of the initial materials based on material similarity data between associated material pairs;

[0008] Determining a material score for each of the candidate materials, and selecting a target material from the candidate materials based on the material score;

[0009] The target material is recommended to the target user.

[0010] In a second aspect, an embodiment of the present application further provides a material recommendation device, comprising:

[0011] An initial material acquisition module is used to select at least one initial material from the material library based on the historical interaction data of the target user;

[0012] a candidate material selection module, configured to select at least one candidate material associated with each of the initial materials from the material library based on material similarity data between associated material pairs;

[0013] A target material selection module is used to determine the material score of each candidate material and select a target material from the candidate materials based on the material score;

[0014] The target material recommendation module is used to recommend the target material to the target user.

[0015] In a third aspect, an embodiment of the present application further provides an electronic device, including:

[0016] one or more processors;

[0017] a memory for storing one or more programs;

[0018] When the one or more programs are executed by the one or more processors, the one or more processors implement any one of the material recommendation methods provided in the embodiment of the first aspect of the present application.

[0019] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any one of the material recommendation methods provided in the embodiment of the first aspect of the present application.

[0020] The embodiment of the present application selects at least one initial material from the material library based on the historical interaction data of the target user; selects at least one candidate material associated with each initial material from the material library based on the material similarity data between the associated material pairs; determines the material score of each candidate material, and selects the target material from the candidate materials based on the material score; and recommends the target material to the target user. The above technical solution determines the candidate materials by accessing the historical interaction data of the target user and the material similarity data between the associated material pairs, and uses the determined candidate materials as the basis for determining the target material, thereby ensuring that the final determined target material can match the user's interaction habits, thereby improving the matching degree between the target material and the target user, and at the same time, avoiding the repeated recommendation of materials that the target user has interacted with, thereby improving the user experience of the target user. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flow chart of a material recommendation method provided in Example 1 of the present application;

[0022] Figure 2 This is a flow chart of another material recommendation method provided in Example 2 of the present application;

[0023] Figure 3A This is a process diagram of a material recommendation method provided in Example 3 of the present application;

[0024] Figure 3B This is a schematic diagram of the structure of a similarity prediction model provided in Example 3 of the present application;

[0025] Figure 4 This is a structural diagram of a material recommendation device provided in Example 4 of the present application;

[0026] Figure 5 This is a structural diagram of an electronic device provided in Example 5 of the present application. DETAILED DESCRIPTION

[0027] The present application will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are intended only to illustrate the present application and are not intended to limit the present application. It should also be noted that, for ease of description, the accompanying drawings only illustrate portions of the present application, not all of the structures.

[0028] Example 1

[0029] Figure 1 This is a flow chart of a material recommendation method provided in an embodiment of the present application. This embodiment is applicable to recommending materials in a material library to target users. Among them, the material can be news, commodities or other messages, etc., and this application does not impose any restrictions on the specific content of the material. The various material recommendation methods provided in this application can be executed by a material recommendation device, which can be implemented in software and / or hardware and specifically configured in an electronic device. In an optional embodiment, the electronic device can be an intelligent terminal or server with certain data processing capabilities.

[0030] See also Figure 1 A material recommendation method shown includes:

[0031] S110 . Select at least one initial material from a material library based on historical interaction data of a target user.

[0032] The material library stores at least one material to be recommended. Each material to be recommended can be a news material, a commodity material, or other information. Different material libraries can be constructed or used according to different recommendation scenarios.

[0033] The historical interaction data may be data generated based on the target user's historical interaction behaviors with each item in the item library, wherein the interaction behaviors may include at least one of browsing, clicking, collecting, sharing, and commenting.

[0034] The material library can be built by users to add, delete and modify materials according to their needs.

[0035] For example, the material library can be pre-stored locally on the electronic device, in other storage devices associated with the electronic device, or in the cloud. Accordingly, when material recommendations are needed, the material library can be searched and used from the local electronic device, in other storage devices associated with the electronic device, or in the cloud.

[0036] It should be noted that the embodiment of the present application does not limit the timing of building the material library. It only needs to ensure that the material library can be used normally before the material recommendation is made this time.

[0037] For example, materials of historical interaction behaviors of target users may be collected, and the collected materials may be used as initial materials.

[0038] It can be understood that by selecting initial materials from the material library based on the target user's historical interaction behavior, each selected initial material can be associated with the target user's behavior, thereby ensuring that the target materials subsequently determined are compatible with the target user. Furthermore, by initially screening the material library based on the user's historical interaction behavior, the recommended material range is narrowed, thereby reducing the amount of data computation required for subsequent target material selection.

[0039] S120. Select at least one candidate material associated with each of the initial materials from the material library according to the material similarity data between the associated material pairs.

[0040] For example, an associated material pair can be a material pair consisting of two materials with an associated relationship in the material library. Material similarity data is used to characterize the similarity attributes between the associated material pairs. A greater similarity indicates a stronger association between the associated material pairs; a smaller similarity indicates a weaker association between the associated material pairs. Therefore, material similarity data can be used to identify similar materials corresponding to a particular material, thereby using these similar materials as the basis for recommending the target material. This ensures consistency between the target material and the target interaction behavior, while preventing the target material from being identified as the same material with which the target user has historically interacted. This, in turn, ensures the subsequent matching between the target material and the target user.

[0041] The material similarity data may be cosine similarity. Of course, those skilled in the art may also use other data representing similarity attributes as material similarity data according to needs or actual conditions, and this application does not impose any limitation on this.

[0042] In an optional embodiment, based on the material similarity data between the associated material pairs, initial materials with similarity values ​​greater than a set similarity threshold can be selected as candidate materials. The set similarity threshold can be determined by technicians based on needs or experience, or can be determined repeatedly through a large number of experiments.

[0043] S130: Determine a material score for each candidate material, and select a target material from the candidate materials based on the material score.

[0044] In an optional embodiment, the material score of each candidate material may be determined by: selecting reference materials corresponding to the candidate material from a material library based on historical interaction data of the target user within a set historical period; and determining the material score of the candidate material based on material similarity data between the candidate material and the reference materials.

[0045] Exemplarily, for each candidate material, materials with which the target user has interacted within a set historical time period are selected from the material library as reference materials; the sum of the material similarity data between the candidate material and the reference materials is calculated, and the sum is used as the material score of the candidate material.

[0046] Specifically, the material score of a candidate material can be determined using the following formula:

[0047] score(user, item1)=∑w item1，item2 r user，item2

[0048] Among them, score(user,item1) represents the material score of candidate material item1 of target user user; w item1,item2 Represents the material similarity data of candidate material item1 relative to each reference material item2; r user,item2 Indicates the target user user's interaction with the material item2 within the set historical time period. If there is any interaction behavior within the set interaction time period, r is 1; otherwise, r is 0.

[0049] It should be noted that the historical time period can be set or adjusted by technical personnel according to needs or experience, and can also be determined through a large number of experiments.

[0050] For example, selecting a target material from candidate materials based on a material score may include selecting at least one candidate material as the target material according to a set selection rule. The set selection rule may include at least one of the following: selecting a material with a score greater than a set score threshold, selecting a quantity less than a set quantity threshold, and selecting a category within a set category range. The set score threshold, set quantity threshold, and set category range may be set or adjusted by technicians or target users based on needs or experience, or determined through extensive testing.

[0051] S140: Recommend the target material to the target user.

[0052] Each target material is sent to a user terminal used by a target user so that the material is displayed through the user terminal.

[0053] The embodiment of the present application selects at least one initial material from the material library based on the historical interaction data of the target user; selects at least one candidate material associated with each initial material from the material library based on the material similarity data between the associated material pairs; determines the material score of each candidate material, and selects the target material from the candidate materials based on the material score; and recommends the target material to the target user. The above technical solution determines the candidate materials by accessing the historical interaction data of the target user and the material similarity data between the associated material pairs, and uses the determined candidate materials as the basis for determining the target material, thereby ensuring that the final determined target material can match the user's interaction habits, thereby improving the matching degree between the target material and the target user, and at the same time, avoiding the repeated recommendation of materials that the target user has interacted with, thereby improving the user experience of the target user.

[0054] Example 2

[0055] Figure 2 This is a flow chart of a material recommendation method provided in an embodiment of the present application. This embodiment is optimized and improved based on the above-mentioned technical solutions.

[0056] Furthermore, before the operation of "selecting at least one candidate material associated with each of the initial materials from the material library based on the material similarity data between the associated material pairs", the operation of "constructing associated material pairs for each material in the material library; wherein the associated material pairs include a first material and a second material; generating associated material word vectors based on the text data of the first material and the second material; determining the similarity of the first material to the second material based on the associated material word vectors; and constructing material similarity data based on the similarities of each associated material pair" is added to improve the mechanism for determining material similarity data between associated material pairs.

[0057] See also Figure 2 A material recommendation method shown includes:

[0058] S210: Construct associated material pairs for each material in the material library; wherein the associated material pairs include a first material and a second material.

[0059] Exemplarily, the associated material pair may be a directed material pair, that is, the associated material pair in which the first material points to the second material is different from the associated material pair in which the second material points to the first material.

[0060] Exemplarily, the associated material pair may also be an undirected material pair or a bidirectional material pair, that is, the associated material pair corresponding to the first material pointing to the second material is the same as the associated material pair corresponding to the second material pointing to the first material.

[0061] S220. Generate an associated material word vector based on the text data of the first material and the second material.

[0062] The text data of the material may include at least one of the material's title data, summary data, introduction data, and detail data.

[0063] It should be noted that, since the title data of a material generally contains summary information of the material, can represent the overall content of the material, and has a small amount of data, in order to reduce the amount of subsequent data calculations, the title data is usually used as the text data of the material.

[0064] Exemplarily, a first material word vector is generated based on the text data of the first material; a second material word vector is generated based on the text data of the second material; the first material word vector and the second material word vector are concatenated in sequence to construct an associated material word vector.

[0065] Optionally, a material word vector is generated based on the text data of the material, which can be: segmenting the text data of the material; encoding each segmentation result separately to obtain a segmentation vector; and concatenating each segmentation vector according to the order of each segmentation result in the text data to obtain the material word vector.

[0066] The encoding process may be implemented based on the Word2vec model or at least one of other models in the prior art, and this application does not impose any limitation on this.

[0067] S230. Determine the similarity between the first material and the second material based on the associated material word vector.

[0068] In an optional embodiment, determining the similarity of the first material to the second material based on the associated material word vector can be: directly determining the cosine similarity between the first material word vector and the second material word vector in the associated material word vector, and using the cosine similarity result as the similarity of the first material to the second material.

[0069] In another optional embodiment, to improve the accuracy of the similarity determination result, the similarity between the first material and the second material can be determined based on the associated material word vectors using a similarity prediction model. The similarity prediction model is trained in the following manner: constructing a sample material pair based on historical interaction data of sample users; wherein the sample material pair includes a first sample material and a second sample material; generating a sample material word vector for the sample material pair based on the text data of the first sample material and the second sample material; and training a pre-constructed machine learning model based on the label data and sample material word vectors of the sample material pair to obtain the similarity prediction model.

[0070] Optionally, the sample material pair can be a directed material pair, such as a sample material pair where the first sample material points to the second sample material, or vice versa. For example, the order of the first and second sample materials can be determined by the chronological order of the interaction behaviors, thereby determining the directional relationship of the directed material pair. For example, the material with the first interaction behavior is designated as the first sample material, and the material with the later interaction behavior is designated as the second sample material.

[0071] Optionally, the sample material pair may also be an undirected material pair or a bidirectional material pair.

[0072] In a specific implementation, to ensure the logic between the constructed sample item pairs, sample item pairs are usually constructed based on interaction habits, with two items constrained by adjacent primary and secondary interaction behaviors. For example, when a user first browses item A and then clicks on item B, browsing is the primary interaction behavior, and clicking is the secondary interaction behavior. In this case, (item A, item B) is constructed as a sample item pair pointing from item A to item B. For another example, when a user first clicks on item B and then adds item C to their favorites, clicking is the primary interaction behavior, and adding to their favorites is the secondary interaction behavior. In this case, (item B, item C) is constructed as a sample item pair pointing from item B to item C.

[0073] It is understandable that in order to ensure the data volume of the sample material pair and avoid the impact of different sample material pair construction methods on subsequent results, the two materials specified when the primary interaction behavior is browsing and the secondary interaction behavior is clicking are usually selected to construct the sample material pair.

[0074] Exemplarily, the pre-built machine learning model can be a bidirectional long short-term memory (LSTM) network or a gated recurrent unit (GRU) network.

[0075] It can be understood that by using a trained similarity prediction model to determine the similarity of the first material to the second material, the associated information in the first material word vector corresponding to the first material and the second material word vector corresponding to the second material can be extracted, and / or irrelevant information corresponding to the first material word vector and the second material word vector can be shielded, thereby improving the accuracy of the similarity judgment result.

[0076] In one optional embodiment, to ensure the generalization ability of the trained similarity prediction model, during model training, the sample material pairs are typically configured to include positive sample material pairs and negative sample material pairs. It is understood that to facilitate sample collection while ensuring the performance of the trained model, a numerical ratio of positive sample material pairs to negative sample material pairs can typically be pre-set. This numerical ratio is generally set by technicians based on needs or experience, or is repeatedly determined through extensive testing. For example, the ratio of positive sample material pairs to negative sample material pairs is 1:4.

[0077] Optionally, constructing a positive sample material pair based on historical interaction data of sample users may be: based on the historical interaction data of sample users, sequentially combining a first sample material and a second sample material that interacted sequentially to construct a positive sample material pair.

[0078] Optionally, constructing a negative sample material pair based on historical interaction data of sample users can be: directly using other sample materials that have no sequential interaction with the first sample material as the third sample material; and sequentially combining the first sample material and the third sample material to construct a negative sample material pair.

[0079] However, simply and mechanically using other sample materials that have no sequential interaction with the first sample material as the third sample material to construct negative sample material pairs results in uneven negative samples, failing to truly reflect the data distribution in the recommendation scenario, and thus affecting the accuracy of the resulting similarity prediction model. To avoid this, the following negative sample material pair construction method can optionally be used to optimize the negative sample material pair construction process: a third sample material is sampled from other sample materials that have no sequential interaction with the first sample material; the first and third sample materials are sequentially combined to construct a negative sample material pair.

[0080] It is understandable that determining the third sample material by sampling can, to a certain extent, avoid constructing negative samples of unseen materials, thereby making the constructed negative sample material pairs more reasonable.

[0081] It should be noted that if the associated item pair is an undirected or bidirectional item pair, the similarity between the first item and the second item and the similarity between the second item and the first item can be averaged. This average value is used as the similarity between the first and second items and used to construct the item similarity data. Because users' primary and secondary interactions follow a sequential order, directed item pairs are typically used to determine the similarity between the first item and the second item, and the item similarity data is then determined based on the similarity between the first item and the second item.

[0082] S240: Construct material similarity data based on the similarity of each associated material pair.

[0083] Exemplarily, the associated material word vectors of the associated material pair are input into a trained similarity prediction model to obtain the similarity of the first material in the associated material pair relative to the second material; based on the correspondence between the material entities involved in the similarity, a similarity matrix is ​​generated, and the similarity matrix is ​​used as the material similarity data.

[0084] Specifically, each first material is used as a row vector, each second material corresponding to the first material is used as a column vector, and the similarity between the first material and the second material is used as an element to generate a similarity matrix, thereby obtaining material similarity data.

[0085] To reduce the amount of data computation, material similarity data is generated for each associated material pair in the material library, and then material recommendations are made based on this material similarity data. To reduce repeated data calculations during the material recommendation process, the generated material similarity data can be pre-stored locally on the electronic device, on other storage devices associated with the electronic device, or in the cloud. When making subsequent material recommendations, the material similarity data can be retrieved and reused.

[0086] Over time, materials in the material library may be updated due to additions and / or deletions. The following details how to update material similarity data when the material library is updated.

[0087] In an optional embodiment, if there is an update caused by deleting a set material in the material library, the material similarity data is traversed, and similarity data related to the deleted material is found, and the found similarity data is deleted, thereby achieving an update of the material similarity data.

[0088] In another optional embodiment, if there is an update caused by adding a new material to the material library, the material similarity data can be updated in the following manner: construct a new material pair with each material in the material library, where the new material pair includes a first new material and a second new material; generate a new material word vector based on the text data of the first new material and the second new material; determine the similarity of the first new material to the second new material based on the new material word vector; and update the material similarity data based on the similarity of each new material pair. In this case, one of the first new material and the second new material is a new material, and the other is an existing material in the material library.

[0089] Exemplarily, a first new material word vector is generated based on the text data of the first new material; a second new material word vector is generated based on the text data of the second new material; the first new material word vector and the second new material word vector are concatenated in sequence to construct a new material word vector.

[0090] The text data of the newly added material may include at least one of the following: title data, summary data, introduction data, and detailed data of the newly added material. It should be noted that the text data of the newly added material can be consistent with the text data used when determining the material similarity data of other materials in the material library.

[0091] S250: Select at least one initial material from a material library based on historical interaction data of the target user.

[0092] S260: Select at least one candidate material associated with each of the initial materials from the material library according to the material similarity data between the associated material pairs.

[0093] S270: Determine a material score for each candidate material, and select a target material from the candidate materials based on the material score.

[0094] S280: Recommend the target material to the target user.

[0095] The embodiment of the present application refines the material similarity data determination process into determining associated material pairs with associated interactive relationships based on the historical interaction data of the recommended user; wherein the associated material pairs include a first material and a second material; generating an associated material word vector based on the text data of the first material and the second material; determining the similarity of the first material to the second material based on the associated material word vector; and constructing material similarity data based on the similarity of each associated material pair. The above technical solution constructs associated material pairs by introducing the historical interaction data of the recommended user, so that the first material and the second material in the constructed material pair can implicitly represent the user behavior, laying the foundation for improving the matching degree between the material recommendation results and the user. The introduction of the text data of the first material and the second material in the associated material pair provides the basis for similarity determination, provides rich and comprehensive data support for the similarity determination process, and improves the accuracy of the similarity determination results. Furthermore, by updating the material similarity data by adding new materials, and then recommending materials based on the updated material similarity data, a rapid cold start of the material is achieved.

[0096] Example 3

[0097] The present application embodiment provides a preferred implementation method based on the above technical solutions. Figure 3AA material recommendation method shown includes an offline phase, an updating phase, and an online phase.

[0098] In an optional embodiment, the offline stage includes several sub-stages: obtaining material titles, labeling positive and negative samples, generating title vectors, and constructing similarity data.

[0099] Exemplarily, obtaining material titles includes: obtaining sample material titles of news materials that have been historically browsed and clicked by sample users.

[0100] Exemplarily, the positive and negative sample marking includes: for the same sample user, if the sample user has an interactive behavior of browsing the first sample material and then clicking on the second sample material, the first sample material and the second sample material are sequentially spliced ​​to generate a positive sample in which the first sample material points to the second sample material; the third sample material is sampled from the materials that have not been clicked, and the first sample material and the third sample material are sequentially spliced ​​to generate a negative sample in which the first sample material points to the third sample material.

[0101] Exemplarily, title vector generation includes: segmenting each material title in the positive and negative samples, and encoding each segmentation result to generate an embedding vector; for each sample, according to the material splicing order and the order of each segmentation result in the material title, splicing each embedding vector to generate a sample title vector corresponding to each sample.

[0102] Exemplarily, similarity data construction includes: inputting sample title vectors into a preset machine learning model, and optimizing the network parameters in the machine learning model according to the positive and negative sample labels corresponding to each sample title vector; using the trained machine learning model as a similarity prediction model. Constructing associated material pairs for each material in the material library; wherein the associated material pairs include a first material and a second material; performing word segmentation and encoding processing on the material title of the first material and the material title of the second material respectively, to obtain an embedding vector; splicing each embedding vector in the order of the first material and the second material and the order in the material title to which they belong, to obtain an associated material word vector; inputting the associated material word vector into the trained similarity prediction model to obtain the similarity of the first material to the second material; using each first material as a row vector, each second material as a column vector, and the similarity of the first material to the second material as a matrix element, to construct a similarity matrix, and using the similarity matrix as the material similarity data.

[0103] In an alternative embodiment, the machine learning model can be implemented based on a long short-term memory network. Figure 3BThe following diagram shows the structure of a similarity prediction model, which includes an embedding layer, a bidirectional LSTM layer, and an output layer. The embedding layer encodes the input word segments to generate word vectors; the bidirectional LSTM layer extracts features from the input word vectors to generate latent vectors; and the output layer activates the latent vectors to generate a probability between 0 and 1, which is used as the similarity between the two items.

[0104] During model training, a loss function (e.g., cross entropy) is determined between the output of the output layer and the positive and negative sample labels, and the network parameter adjustment process of the preset machine learning model is guided by the loss function until the training cutoff condition is met. For example, at least one of the following conditions is met: the number of total training samples meets a set quantity threshold, the numerical result of the loss function tends to be stable, or the function value of the loss function is less than a set numerical threshold. The size of the set quantity threshold and the set numerical threshold can be determined or adjusted by the technician according to needs or experience.

[0105] It should be noted that, in the offline stage, considering the real-time nature of news recommendations, the material library is usually updated on a daily basis at the latest; if the number of news items is large, the material library can also be updated on an hourly basis.

[0106] In an optional embodiment, when there are new materials in the material library, the material recommendation process also includes an update stage.

[0107] Exemplarily, the update stage may include: construction of new material pairs: obtaining the title data of the new material, and constructing new material pairs for the new material and other materials in the material library; generation of new word vectors: segmenting, encoding and splicing each material in the new material pair to obtain the new word vector corresponding to the new material pair; similarity data update: based on the trained similarity prediction model, determining the new similarity between the materials in the new material pair corresponding to the new word vector, and updating the material similarity data according to the new similarity.

[0108] In an optional embodiment, the online stage includes several sub-stages: material screening, material expansion, score calculation and material recommendation.

[0109] Exemplarily, material screening includes: obtaining initial materials that the user to be recommended has historically interacted with in the material library.

[0110] Exemplarily, material expansion includes: based on the material similarity data, screening a first set number of other materials from the material library whose similarity to each initial material meets a set similarity threshold as candidate materials. The similarity threshold value and the value of the first set number can be determined or adjusted by technical personnel based on needs or experience. Optionally, the first set number can be related to the number or type of initial materials. For example, if there are three initial materials, the set number is 3K, where K is the set sub-number corresponding to each initial material.

[0111] For example, if user A has historically interacted with materials A, B, and C, then based on the material similarity data, K materials A' whose similarity with A exceeds the set similarity threshold are screened out from the material library, K materials B' whose similarity with B exceeds the set similarity threshold are screened out, and K materials C' whose similarity with C exceeds the set similarity threshold are screened out; A', B', and C' are taken as candidate materials.

[0112] For example, the score calculation may be: using the following formula to calculate the material score corresponding to each candidate material:

[0113] score(user, item1)=∑w item1，item2 r user，item2

[0114] Among them, score(user,item1) represents the material score of candidate material item1 of target user user; w item1,item2 Represents the material similarity data of candidate material item1 relative to each reference material item2 (other materials in the material library); r user,item2 Indicates the target user user's interaction with the material item2 within the set historical time period. If there is any interaction behavior within the set interaction time period, r is 1; otherwise, r is 0.

[0115] Exemplarily, material recommendation includes selecting a second set number of candidate materials with the highest material scores from among the candidate materials as target materials, and sending the target materials to the user terminal of the user to whom the recommendation is to be made for display. The second set number can be set by technical personnel based on needs or experience, or determined or adjusted through extensive testing. The second set number is smaller than the first set number.

[0116] Example 4

[0117] Figure 4This is a structural diagram of a material recommendation device provided in an embodiment of the present application. The device is suitable for recommending materials from a material library to a target user. The materials may be news, products, or other information, and this application does not impose any restrictions on the specific content of the materials. The device can be implemented using software and / or hardware and specifically configured in an electronic device. In an optional embodiment, the electronic device can be an intelligent terminal or server with certain data processing capabilities.

[0118] See also Figure 4 The material recommendation device shown includes: an initial material acquisition module 410, a candidate material selection module 420, a target material selection module 430 and a target material recommendation module 440.

[0119] An initial material acquisition module 410 is configured to select at least one initial material from a material library based on historical interaction data of a target user;

[0120] a candidate material selection module 420 for selecting at least one candidate material associated with each of the initial materials from the material library based on material similarity data between associated material pairs;

[0121] a target material selection module 430 for determining a material score of each candidate material and selecting a target material from the candidate materials based on the material score;

[0122] The target material recommendation module 440 is configured to recommend the target material to the target user.

[0123] The embodiment of the present application selects at least one initial material from the material library based on the historical interaction data of the target user; selects at least one candidate material associated with each initial material from the material library based on the material similarity data between the associated material pairs; determines the material score of each candidate material, and selects the target material from the candidate materials based on the material score; and recommends the target material to the target user. The above technical solution determines the candidate materials by accessing the historical interaction data of the target user and the material similarity data between the associated material pairs, and uses the determined candidate materials as the basis for determining the target material, thereby ensuring that the final determined target material can match the user's interaction habits, thereby improving the matching degree between the target material and the target user, and at the same time, avoiding the repeated recommendation of materials that the target user has interacted with, thereby improving the user experience of the target user.

[0124] In an optional embodiment, the apparatus further comprises: a material similarity data determination module, configured to determine material similarity data between associated material pairs;

[0125] The material similarity data determination module specifically includes:

[0126] an associated material pair determining unit, configured to construct associated material pairs for each material in the material library; wherein the associated material pairs include a first material and a second material;

[0127] an associated material word vector generating unit, configured to generate an associated material word vector based on the text data of the first material and the second material;

[0128] a similarity determination unit, configured to determine a similarity between the first material and the second material based on the associated material word vector;

[0129] The similarity data construction unit is used to construct material similarity data according to the similarity of each associated material pair.

[0130] In an optional embodiment, the similarity determination unit includes:

[0131] a similarity determination subunit, configured to determine the similarity between the first material and the second material based on the associated material word vector using a similarity prediction model;

[0132] Wherein, the similarity determination unit further includes: a similarity model training subunit, for training the similarity prediction model;

[0133] The similarity model training subunit is specifically used to:

[0134] Constructing a sample material pair based on historical interaction data of sample users; wherein the sample material pair includes a first sample material and a second sample material;

[0135] Generating a sample material word vector for the sample material pair according to the text data of the first sample material and the second sample material;

[0136] The pre-built machine learning model is trained according to the label data of the sample material pair and the sample material word vector to obtain the similarity prediction model.

[0137] In an optional embodiment, the sample material pairs include a positive sample material pair and a negative sample material pair;

[0138] Accordingly, constructing sample material pairs based on historical interaction data of sample users includes:

[0139] According to the historical interaction data of the sample user, sequentially combine the first sample material and the second sample material that interacted sequentially to construct a positive sample material pair;

[0140] Sampling a third sample material from other sample materials that have no sequential interaction with the first sample material;

[0141] The first sample material and the third sample material are sequentially combined to construct a negative sample material pair.

[0142] In an optional embodiment, the device further includes a material similarity data updating module, specifically including:

[0143] A new material pair construction unit is used to construct a new material pair with each material in the material library; wherein the new material pair includes a first new material and a second new material;

[0144] a new material word vector generating unit, configured to generate a new material word vector based on the text data of the first new material and the second new material;

[0145] a similarity determination unit, configured to determine a similarity between the first newly added material and the second newly added material based on the newly added material word vector;

[0146] The material similarity data updating unit is used to update the material similarity data according to the similarity of each newly added material pair.

[0147] In an optional embodiment, the target material selection module 430 includes:

[0148] a reference material selection unit, configured to select reference materials corresponding to the candidate materials from the material library based on the historical interaction data of the target user within a set historical period;

[0149] The material score determining unit is configured to determine the material score of the candidate material based on the material similarity data between the candidate material and each of the reference materials.

[0150] In an optional embodiment, each material in the material library is news.

[0151] The above-mentioned material recommendation device can execute the material recommendation method provided in any embodiment of the present application, and has the corresponding functional modules and beneficial effects for executing the material recommendation method.

[0152] Example 5

[0153] Figure 5 This is a structural diagram of an electronic device provided in Example 5 of the present invention. Figure 5 A block diagram of an exemplary electronic device 512 suitable for implementing embodiments of the present invention is shown. Figure 5 The electronic device 512 shown is only an example and should not limit the functionality and scope of use of the embodiments of the present invention.

[0154] like Figure 5As shown, electronic device 512 is implemented as a general-purpose computing device. Components of electronic device 512 may include, but are not limited to, one or more processors or processing units 516, system memory 528, and a bus 518 that connects various system components (including system memory 528 and processing unit 516).

[0155] Bus 518 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus architectures. Examples of these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.

[0156] The electronic device 512 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the electronic device 512, including volatile and non-volatile media, removable and non-removable media.

[0157] The system memory 528 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 530 and / or cache memory 532. The electronic device 512 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 534 may be used to read and write non-removable, non-volatile magnetic media ( Figure 5 Not shown, often called a "hard drive"). Although Figure 5 Not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 518 via one or more data media interfaces. Memory 528 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of various embodiments of the present invention.

[0158] A program / utility 540 having a set (at least one) of program modules 542 may be stored, for example, in memory 528. Such program modules 542 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data, each of which, or some combination thereof, may include an implementation of a network environment. Program modules 542 generally implement the functions and / or methodologies of the embodiments described herein.

[0159] The electronic device 512 can also communicate with one or more external devices 514 (e.g., a keyboard, pointing device, display 524, etc.), one or more devices that enable a user to interact with the electronic device 512, and / or any device that enables the electronic device 512 to communicate with one or more other computing devices (e.g., a network card, a modem, etc.). Such communication can occur via an input / output (I / O) interface 522. Furthermore, the electronic device 512 can communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network such as the Internet) via a network adapter 520. As shown, the network adapter 520 communicates with other modules of the electronic device 512 via a bus 518. It should be understood that, although not shown, other hardware and / or software modules can be used in conjunction with the electronic device 512, including but not limited to microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0160] The processing unit 516 executes various functional applications and data processing by running at least one of the other programs among the multiple programs stored in the system memory 528, such as implementing the material recommendation method provided by the embodiment of the present invention.

[0161] Example 6

[0162] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, a material recommendation method provided by any embodiment of the present invention is implemented, including: selecting at least one initial material from a material library based on historical interaction data of a target user; selecting at least one candidate material associated with each of the initial materials from the material library based on material similarity data between associated material pairs; determining a material score for each of the candidate materials, and selecting a target material from the candidate materials based on the material score; and recommending the target material to the target user.

[0163] Note that the above are only preferred embodiments of the present application and the technical principles employed. Those skilled in the art will understand that the present application is not limited to the specific embodiments described herein, and that various obvious changes, readjustments, and substitutions can be made by those skilled in the art without departing from the scope of protection of the present application. Therefore, although the present application has been described in more detail through the above embodiments, the present application is not limited to the above embodiments and may include many other equivalent embodiments without departing from the scope of the present application. The scope of the present application is determined by the scope of the appended claims.

Claims

1. A material recommendation method, characterized in that: include: Select at least one initial material from the material library based on the historical interaction data of the target user; selecting, from the material library, at least one candidate material associated with each of the initial materials based on material similarity data between associated material pairs; Determining a material score for each of the candidate materials, and selecting a target material from the candidate materials based on the material score; Determining the material score of each candidate material includes: selecting reference materials corresponding to the candidate material from the material library based on the historical interaction data of the target user within a set historical period; and determining the material score of the candidate material based on the material similarity data between the candidate material and the reference materials. The material score of the candidate material is determined using the following formula: score(user,item1)=∑W item1,item r user,item2 Among them, score(user, item1) represents the material score of candidate material item1 of target user user; W item1,item Represents the material similarity data of candidate material item1 relative to each reference material item2; r user,item2 Indicates the interaction of the target user user with the material item2 within the set historical time period. If an interaction behavior has occurred within the set interaction time period, r is 1; otherwise, r is 0; the material similarity data is a similarity matrix, and the elements in the similarity matrix are generated using a similarity prediction model. The similarity prediction model is trained in the following way: based on the historical interaction data of the sample user, a sample set including positive sample material pairs and negative sample material pairs is constructed, wherein the positive sample material pair is composed of the first sample material and the second sample material that interact sequentially, and the negative sample material pair is constructed by sampling the third sample material from other sample materials that have no sequential interaction with the first sample material; a sample material word vector is generated based on the text data of the sample material pair, and a machine learning model is trained based on the label data; when a new material is added to the material library, a new material pair is constructed between the new material and each material in the material library, a new material word vector is generated, and the material similarity data is updated; The target material is recommended to the target user.

2. The method according to claim 1, characterized in that The material similarity data between the associated material pairs is determined in the following manner: Constructing associated material pairs for each material in the material library; wherein the associated material pairs include a first material and a second material; Generate associated material word vectors based on the text data of the first material and the second material; Determining, based on the associated material word vector, a similarity between the first material and the second material; Material similarity data is constructed based on the similarity of each related material pair.

3. The method according to claim 2, characterized in that The determining, based on the associated material word vector, the similarity between the first material and the second material includes: Use the similarity prediction model to determine the similarity between the first material and the second material based on the associated material word vector 4. The method according to claim 2, characterized in that The material similarity data is updated in the following way: Constructing a new material pair with each material in the material library; wherein the new material pair includes a first new material and a second new material; Generate a new material word vector based on the text data of the first new material and the second new material; Determining, based on the newly added material word vector, a similarity between the first newly added material and the second newly added material; The material similarity data is updated according to the similarity of each newly added material pair.

5. The method according to any one of claims 1 to 4, characterized in that Each material in the material library is news.

6. A material recommendation device, characterized in that: include: An initial material acquisition module is used to select at least one initial material from the material library based on the historical interaction data of the target user; a candidate material selection module, configured to select at least one candidate material associated with each of the initial materials from the material library based on material similarity data between associated material pairs; A target material selection module is used to determine the material score of each candidate material and select a target material from the candidate materials based on the material score; Determining the material score of each candidate material includes: selecting reference materials corresponding to the candidate material from the material library based on the historical interaction data of the target user within a set historical period; and determining the material score of the candidate material based on the material similarity data between the candidate material and the reference materials. The material score of the candidate material is determined using the following formula: score(user,item1)=∑W item1,item r user,item2 Among them, score(user, item1) represents the material score of candidate material item1 of target user user; W item1,item Represents the material similarity data of candidate material item1 relative to each reference material item2; r user,item2 Indicates the interaction of the target user user with the material item2 within the set historical time period. If an interaction behavior has occurred within the set interaction time period, r is 1; otherwise, r is 0; the material similarity data is a similarity matrix, and the elements in the similarity matrix are generated using a similarity prediction model. The similarity prediction model is trained in the following way: based on the historical interaction data of the sample user, a sample set including positive sample material pairs and negative sample material pairs is constructed, wherein the positive sample material pair is composed of the first sample material and the second sample material that interact sequentially, and the negative sample material pair is constructed by sampling the third sample material from other sample materials that have no sequential interaction with the first sample material; a sample material word vector is generated based on the text data of the sample material pair, and a machine learning model is trained based on the label data; when a new material is added to the material library, a new material pair is constructed between the new material and each material in the material library, a new material word vector is generated, and the material similarity data is updated; The target material recommendation module is used to recommend the target material to the target user.

7. An electronic device, characterized in that: include: one or more processors; a memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement a material recommendation method as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, a material recommendation method according to any one of claims 1 to 5 is implemented.

Images