Information recommendation method, device and storage medium

Abstract

The present application discloses an information recommendation method, device and storage medium, which relate to the field of information processing technology and can be applied to various scenarios such as cloud technology, artificial intelligence, smart transportation, and Internet of Vehicles. The method includes: obtaining first target object data of a target object and target association information corresponding to the first target object data; determining second target object data for publishing target association information; determining a second object semantic data set based on the target association information; determining a second object behavior data set based on the first target object data and the second target object data; determining a second object data set to be recommended based on the second object semantic data set and the second object behavior data set; and recommending the association information of the second object data to be recommended to the target object based on the second object data to be recommended. The present application makes recommendations at the granularity of the second object data, thereby recommending information of interest migration to users and increasing the amount of recommended information.

Description

Technical Field

[0001] The present application relates to the field of information processing technology, and in particular to an information recommendation method, device, and storage medium. Background Art

[0002] Existing technologies for information recommendation require manual labeling of information categories. Due to the large volume of information, manual labeling is impossible, and often results in labeling high-profile information, making it difficult for low-profile information to be recommended. Furthermore, manual labeling of information categories is highly subjective and inaccurate, resulting in inaccurate recommendations for users and impacting the exposure of recommended information. Furthermore, existing information recommendation technologies fail to consider interest migration, which reduces the number of recommended information.

[0003] Therefore, it is necessary to provide an information recommendation method, device and storage medium, which can recommend information of interest migration to users and increase the amount of recommended information. Summary of the Invention

[0004] The present application provides an information recommendation method, device and storage medium, which can recommend information of interest migration to users and increase the amount of recommended information.

[0005] In one aspect, the present application provides an information recommendation method, the method comprising:

[0006] Acquire first target object data of a target object and target association information corresponding to the first target object data;

[0007] Determine the second target object data for publishing the target association information;

[0008] Determining, based on the target association information, a second object semantic data set having a semantic similarity with the second target object data greater than a first preset threshold;

[0009] Determine, based on the first target object data and the second target object data, a second object behavior data set having a behavior similarity with the second target object data greater than a second preset threshold;

[0010] determining second object data to be recommended based on the second object semantic dataset and the second object behavior dataset;

[0011] Based on the second object data to be recommended, associated information of the second object data to be recommended is recommended to the target object.

[0012] Another aspect provides an information recommendation device, the device comprising:

[0013] a target association information acquisition module, configured to acquire first target object data of a target object and target association information corresponding to the first target object data;

[0014] A second target object data determining module, configured to determine the second target object data for publishing the target association information;

[0015] A second object semantic data set determining module, configured to determine, based on the target association information, a second object semantic data set having a semantic similarity with the second target object data that is greater than a first preset threshold;

[0016] A second object behavior data set determining module, configured to determine, based on the first target object data and the second target object data, a second object behavior data set having a behavior similarity with the second target object data that is greater than a second preset threshold;

[0017] a second object data to be recommended determining module, configured to determine the second object data to be recommended based on the second object semantic data set and the second object behavior data set;

[0018] The associated information determining module is configured to recommend the associated information of the second object data to be recommended to the target object based on the second object data to be recommended.

[0019] On the other hand, an information recommendation device is provided, which includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the information recommendation method described above.

[0020] On the other hand, a computer storage medium is provided, wherein the computer storage medium stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by a processor to implement the information recommendation method described above.

[0021] Another aspect provides a computer program product or computer program, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to implement the information recommendation method described above.

[0022] The information recommendation method, device, and storage medium provided in this application have the following technical effects:

[0023] The present application obtains first target object data of a target object and target association information corresponding to the first target object data; and determines second target object data for publishing the target association information; based on the target association information, determines a second object semantic data set whose semantic similarity with the second target object data is greater than a first preset threshold; based on the first target object data and the second target object data, determines a second object behavior data set whose behavior similarity with the second target object data is greater than a second preset threshold; based on the second object semantic data set and the second object behavior data set, determines the second object data to be recommended; not only the second object semantic data set that is semantically similar to the second target object data is obtained, but also the second object behavior data set of interest migration is obtained, thereby increasing the number of second object data to be recommended; based on the second object data to be recommended, recommends the association information of the second object data to be recommended to the target object; realizes the recall of the association information to be recommended from the second object data granularity, and increases the number of recommended information. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0025] Figure 1 is a schematic diagram of an information recommendation system provided by an embodiment of the present application;

[0026] Figure 2 This is a flowchart of an information recommendation method provided by an embodiment of the present application;

[0027] Figure 3 This is a flowchart of a method for obtaining first target object data of a target object and target association information corresponding to the first target object data, provided by an embodiment of the present application;

[0028] Figure 4 1 is a flowchart of a method for determining a second object semantic data set having a semantic similarity with the second target object data greater than a first preset threshold value, provided by an embodiment of the present application;

[0029] Figure 5 1 is a flow chart of a method for determining a second object behavior data set having a behavior similarity with the second target object data greater than a second preset threshold value, provided by an embodiment of the present application;

[0030] Figure 61 is a flow chart of a method for determining second object data to be recommended provided in an embodiment of the present application;

[0031] Figure 7 is a flowchart of another method for determining second object data to be recommended provided by an embodiment of the present application;

[0032] Figure 8 Schematic diagram of the Word2Vec model provided in the embodiment of the present application;

[0033] Figure 9 It is a structural diagram of the CBOW model provided in the embodiment of the present application;

[0034] Figure 10 This is a schematic diagram of the Node2Vec graph network structure of the CP provided in an embodiment of the present application;

[0035] Figure 11 It is the association information corresponding to the second target object data provided in an embodiment of the present application;

[0036] Figure 12 is the association information corresponding to the semantic data of the second object to be recommended provided in an embodiment of the present application;

[0037] Figure 13 is another type of associated information corresponding to the second target object data provided in an embodiment of the present application;

[0038] Figure 14 It is the association information corresponding to the second object behavior data to be recommended provided in an embodiment of the present application;

[0039] Figure 15 This is a display interface for recommendation information provided by an embodiment of the present application;

[0040] Figure 16 This is a structural diagram of an information recommendation device provided in an embodiment of the present application;

[0041] Figure 17 This is a structural diagram of a server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0042] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0043] It should be noted that the terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0044] See also Figure 1 , Figure 1 is a schematic diagram of an information recommendation system provided in an embodiment of the present application, such as Figure 1 As shown, the information recommendation system may include at least a server 01 and a client 02 .

[0045] Specifically, in an embodiment of the present application, the server 01 may include an independently operated server, or a distributed server, or a server cluster composed of multiple servers. It may also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Server 01 may include a network communication unit, a processor, a memory, and the like. Specifically, the server 01 can be used to determine the second object data to be recommended based on the second target object data corresponding to the target object, thereby determining the associated information of the second object data to be recommended.

[0046] Specifically, in the embodiment of the present application, the client 02 may include a physical device such as a smartphone, desktop computer, tablet computer, laptop computer, digital assistant, smart wearable device, smart speaker, in-vehicle terminal, smart TV, etc., or may include software running on the physical device, such as a web page provided by a service provider to a user, or an application provided by the service provider to a user. Specifically, the client 02 may be used to display associated information of the second object data to be recommended.

[0047] The following describes an information recommendation method for this application. Figure 2It is a flow chart of an information recommendation method provided by an embodiment of the present application. This specification provides the method operation steps as described in the embodiment or flow chart, but may include more or fewer operation steps based on conventional or non-creative labor. The order of steps listed in the embodiment is only one way of executing the order of many steps and does not represent the only execution order. When the actual system or server product is executed, it can be executed in sequence or in parallel according to the method shown in the embodiment or the accompanying drawings (for example, in a parallel processor or multi-threaded processing environment).

[0048] Specific examples Figure 2 As shown, the method may include:

[0049] S201: Acquire first target object data of a target object and target association information corresponding to the first target object data.

[0050] In the embodiments of the present application, the target object may include but is not limited to a user, a smart device, etc.; the target operation behavior may include but is not limited to the user's click, slide, drag and other behaviors in the terminal display interface; the first target object data may be data formed by the target object's operation behavior; for example, when the target operation behavior is a click behavior, the first target object data may be target operation behavior data, such as the number of clicks, click time, and other data; the first target object data may be data corresponding to a single operation behavior of the target object or multiple consecutive operation behaviors; the target association information is information triggered by the target operation behavior; for example, the target association information may be the article text displayed after the user clicks an article link, or it may be the video content played after the user clicks a video link. The target association information is information in the target application.

[0051] In the embodiments of this application, Figure 3 As shown, the above-mentioned acquisition of the first target object data of the target object and the target association information corresponding to the above-mentioned first target object data includes:

[0052] S20101: Acquire multiple first object data of the target object and multiple candidate association information corresponding to the multiple first object data;

[0053] In an embodiment of the present application, multiple first object data may be data corresponding to multiple operation behaviors of the target object within a preset time period; wherein, each operation behavior corresponds to a candidate association information, for example, when a user clicks on an article link, he can view the article corresponding to the link (candidate association information).

[0054] S20103: Obtaining the display duration of each candidate association information;

[0055] Specifically, in the embodiment of the present application, the display duration of the candidate associated information may be the viewing duration or reading duration of the candidate associated information by the target object.

[0056] S20105: Determine the candidate related information whose display duration is greater than a preset duration threshold as the target related information;

[0057] Specifically, in an embodiment of the present application, when the target object views related information through operational behavior, an accidental click may occur, or the target object may click on related information that it is not interested in and quickly close it; for example, when the user is reading an article or watching a video, the target object may accidentally click on or click on an article or video that it is not interested in; at this time, by setting a preset time threshold, the related information that the target object accidentally clicks on or does not like can be screened out, and the target related information that the target object is interested in can be determined.

[0058] S20107: Determine the first object data corresponding to the target association information as the first target object data.

[0059] In the embodiment of the present application, the first object data corresponding to the target association information that the target object is interested in may be determined as the first target object data.

[0060] In an embodiment of the present application, in the process of obtaining the first object data and target associated information, the conditions for effective clicks can be added. The first target object data and target associated information can be screened out by the display duration of the candidate associated information corresponding to each first object data, that is, the effective first object data and effective associated information are screened out, thereby improving the accuracy of subsequent associated information to be recommended and increasing the exposure of the associated information to be recommended.

[0061] S203: Determine the second target object data for publishing the target association information.

[0062] In an embodiment of the present application, the second target object data may be a publisher identifier, which may include but is not limited to the author identifier of the target-related information, the identifier of the publishing platform, or the content provider identifier (CP). For a news application, the content provider is the publishing media. In this case, the content provider identifier is the media ID.

[0063] S205: Based on the target association information, determine a second object semantic data set whose semantic similarity with the second target object data is greater than a first preset threshold.

[0064] In an embodiment of the present application, the target association information may be information generated based on the first target object data. For example, when the first target object data is target object operation behavior data, the target association information may be data corresponding to the operation behavior. For example, when a user clicks on an article title, the article content that is opened is the corresponding target association information. The target association information may be used to retrieve a plurality of second object semantic data having a high semantic similarity to the second target object data to form a second object semantic data set. The first preset threshold may be set based on actual circumstances. The second object semantic data set may include at least one second object semantic data.

[0065] In embodiments of the present application, target attribute information can also be determined based on the target association information, thereby determining the second object semantic dataset based on the target attribute information. The target attribute information may include, but is not limited to, the title, type, category, and second target object data of the target association information. For example, if the target association information is an article, the target attribute information may include, but is not limited to, the article title, type tag, first-level category, and second-level category.

[0066] In an embodiment of the present application, the above method may further include:

[0067] Target words within a preset time period are obtained, where the target words represent words that have been searched more than a preset threshold number of times.

[0068] Specifically, in an embodiment of the present application, the target vocabulary may be a hot word within a preset time period, such as a word on a hot search list in a browser App.

[0069] In the embodiment of the present application, determining, based on the target association information, a second object semantic dataset having a semantic similarity with the second target object data greater than a first preset threshold may include:

[0070] Based on the target association information and the target vocabulary, a second object semantic data set is determined, the second object semantic data set having a semantic similarity with the second target object data greater than a first preset threshold.

[0071] In the embodiment of the present application, the target association information and the target vocabulary may be combined to determine the second object semantic data set, thereby improving the click-through rate and exposure of the recommended information.

[0072] In the embodiments of this application, Figure 4 As shown, the above-mentioned determining, based on the above-mentioned target association information, a second object semantic data set having a semantic similarity with the above-mentioned second target object data greater than a first preset threshold includes:

[0073] S2051: Extracting semantic features from the target association information to obtain target semantic features of the second target object data;

[0074] In an embodiment of the present application, semantic features can be extracted from the target association information based on a semantic feature extraction model to obtain the target semantic features of the above-mentioned second target object data; the above-mentioned semantic feature extraction model is trained based on the sample association information and sample semantic features of the sample object.

[0075] In a specific embodiment, in CP recommendation, manual labeling of CP categories can be used to label CP post types, such as (basketball CP, entertainment gossip CP), and the semantics of CP can be determined through the content of the post. This solution has three disadvantages. First, manpower cannot meet the needs of millions of CPs, and can only label high-exposure CPs. New CPs are difficult to label and cannot participate in downstream recommendations; second, manual labeling of CP semantics is to label the content of the current CP post, usually only once, and the posts after the CP will not be recorded, and routine updates cannot be achieved. The CP semantics used downstream are usually lagging and inaccurate. Third: A large amount of manual labeling cannot make horizontal comparisons of CPs like a machine, and the labeling will be biased.

[0076] In an embodiment of the present application, a machine model can be used to learn the semantic vector representation (Embedding) of CP, and all news CPs can be used to routinely update the CP semantic Embedding to achieve machine labeling of all CPs in the application, and update them daily to ensure the accuracy of the recommended information.

[0077] In an embodiment of the present application, the training method of the above-mentioned semantic feature extraction model may include:

[0078] Get sample-related information of the sample object;

[0079] Determining second sample object data for publishing sample-related information, and marking sample semantic features corresponding to the second sample object data;

[0080] Conduct semantic feature extraction training based on machine learning models;

[0081] Adjusting parameters of the machine learning model until the output semantic features match the input second sample object data;

[0082] The machine learning model corresponding to the model parameters when the output semantic feature matches the input second sample object data is used as the semantic feature extraction model.

[0083] Specifically, in an embodiment of the present application, in a news application, sample association information may include news post content, such as the title, primary category, CP, secondary category, and tags in a post; and the title is segmented; in a specific embodiment, the obtained sample association information may be: (marketing, master's path, use, promotion, product, software, acquisition, traffic, skills, technology, software system, CP1, Jibaotai, online marketing, web page), where CP is regarded as a word, and all the features of a post are combined into a sentence.

[0084] Specifically, in an embodiment of the present application, the machine learning model can be an unsupervised model Word2Vec, which is a group of related models used to generate word vectors. These models are shallow, two-layer neural networks used for training to reconstruct linguistic word texts. The network is represented by words, and it is necessary to guess the input words in adjacent positions. Under the word bag model assumption in Word2vec, the order of words is not important. After training is completed, the Word2vec model can be used to map each word to a vector, which can be used to represent the relationship between words. The vector is the hidden layer of the neural network. Word2Vec is essentially a neural network with only one hidden layer, and its corresponding structure is as follows. Figure 8 As shown in the figure, semantic embedding uses the Continuous Bag-Of-Words (CBOW) model in Word2Vec, where CBOW predicts the target word based on the context, with the goal of maximizing the probability of occurrence of the target word, as shown in the figure. Figure 9 As shown in the figure, assume that the vocabulary size is V, the word vector dimension is N, and the context words are x1, x2, ..., CP, ..., xC. Here, x1, x2, xC are words obtained by segmenting based on the attribute information of the associated information. Define the context window size as C, where C represents the total number of word vectors before and after CP. The corresponding target word is y. First, represent both x and y in one-hot encoding form. One-hot encoding is the process of converting categorical variables into a form that is easier to use in machine learning algorithms. CP is usually placed in the middle of the word, but it is not limited to the middle; its position can vary.

[0085] There are two matrix parameters involved here. W is the word vector matrix, where each row is the word vector v for a certain word. W' can be seen as an auxiliary matrix, where each column can be seen as the corresponding vector v' for a certain word. For example, N can be 128 and v can be 256.

[0086] Forward process: Input layer to hidden layer: For each word xi, extract the corresponding word vector vi, then take the average of these word vectors as the hidden word vector to obtain the semantic features of these words. Hidden layer to output layer: Multiply h by W' to obtain a vector u of dimension V, normalize it through the normalized exponential function (softmax) to obtain a probability vector, and take the one with the highest probability as the predicted word.

[0087] Backpropagation process: The model needs to maximize the probability of the target word, that is, to minimize the negative log-likelihood function. The model needs to update two matrix parameters, W and W', calculate the gradient of the parameters according to the loss function, and then use the gradient descent method to update the parameters.

[0088] After model training converges, each word (including each CP) will have a feature vector, where similar words will cluster in similar vector spaces, thereby learning the relationship between words and words and CPs. The semantic embedding of the CP will carry the semantics of the post content.

[0089] In an embodiment of the present application, the target semantic features of the second target object data can be quickly determined based on the semantic feature extraction model, thereby facilitating the subsequent rapid recall of the candidate semantic feature set.

[0090] S2053: Determine, based on a first preset database, a set of candidate semantic features whose similarity to the target semantic feature is greater than the first preset threshold; the first preset database stores the semantic features corresponding to the second preset object data;

[0091] In an embodiment of the present application, the first preset database can be updated regularly, and the update time can be set according to actual needs, for example, it can be set to update once a day. By regularly updating, the accuracy of the candidate semantic specific set can be improved, thereby improving the accuracy of the recommended information.

[0092] In the embodiment of the present application, the similarity between two features can be determined based on the cosine distance between the two semantic feature vectors;

[0093] S2055: Obtain a candidate semantic second object data set based on the candidate semantic second object data corresponding to each candidate semantic feature in the candidate semantic feature set.

[0094] In an embodiment of the present application, in order to facilitate experiments on the recommendation system and to increase the speed in the similarity search process, the CP semantic Embedding sets the vector dimension to 32 and the sliding window to 5. The Embedding similarity retrieval method is used to output to the recommendation system the top 200 CPs with the closest semantics to each CP in the total number of CPs; that is, the candidate semantic second object data set can include 200 candidate semantic second object data.

[0095] S207: Based on the first target object data and the second target object data, determine a candidate behavior second object data set having a behavior similarity with the second target object data that is greater than a second preset threshold.

[0096] In the embodiments of this application, Figure 5 As shown, the above-mentioned determining, based on the above-mentioned first target object data and the above-mentioned second target object data, a candidate behavior second object data set having a behavior similarity with the above-mentioned second target object data greater than a second preset threshold includes:

[0097] S2071: extracting behavioral features from the first target object data and the second target object data based on a behavioral feature extraction model to obtain target behavioral features of the second target object data;

[0098] In this embodiment of the application, the second object data is coarser in granularity than the article and richer in content. Compared to existing semantic recommendation, this embodiment learns the relationship between the second object data from online user behavior and uses it as a new recall branch to increase effective recall.

[0099] In an embodiment of the present application, the above method further includes: a training method for a behavior feature extraction model, and the above training method for a behavior feature extraction model includes:

[0100] S1. Obtain a second sample object data set, where each second sample object data in the second sample object data set is marked with a corresponding behavior feature label.

[0101] S2. Take each second sample object data in the above second sample object data set as a node, connect any two second sample object data that are clicked continuously as an edge, and use the number of times the above any two second sample object data are clicked continuously as the weight of the corresponding edge to construct a second sample object data sequence.

[0102] In this embodiment of the present application, the corpus is constructed by mining a user's online logs, mapping a user's click sequence to articles, and obtaining a CP sequence, i.e., a second sample object data sequence, such as (CP1-CP2-CP3-CP4). The CP sequence can be a sequence corresponding to a user's click behavior within a preset time period, for example, a CP sequence corresponding to a user's continuous click behavior within 1 minute.

[0103] In the embodiment of the present application, the graph network model Node2Vec can be used. The idea of Node2vec is the same as the random walk algorithm: generate random walks, sample the random walks to obtain a combination of (nodes, contexts), and then use the word vector processing method to model such a combination to obtain the representation of the network node. Figure 10 As shown in the figure, CP is used as a node, and continuous clicks on CP connect two CPs as edges. The number of continuous clicks is used as the weight of the edge. For example, for CP1-CP2, two users clicked on these two CPs continuously, and the weight of the corresponding edge is 2. Sampling is performed by random walk, and a corresponding sequence is generated for each point.

[0104] S3. Based on each second sample object data in the above-mentioned second sample object data sequence, the preset machine learning model is trained for behavioral feature extraction to adjust the model parameters of the above-mentioned preset machine learning model until the behavioral feature label of each second sample object data output by the above-mentioned preset machine learning model matches the marked behavioral feature label.

[0105] In an embodiment of the present application, the second sample object data sequence can be regarded as text imported into the CBOW model in Word2Vec to obtain the vector of each node (corresponding to the vector of each word in Word2Vec), where the node is CP and the final output is the feature vector of each CP.

[0106] S4. The machine learning model corresponding to the model parameters when the output behavior feature label matches the labeled behavior feature label is used as the above-mentioned behavior feature extraction model.

[0107] In this embodiment of the present application, based on the constructed second sample object data sequence, it can be seen that users who like to read entertainment news also pay attention to movie news; those who like to watch food programs also watch parenting reports; therefore, the user's interest migration behavior can be determined. The behavioral feature extraction model considers the user's migration behavior when extracting behavioral features. Therefore, the subsequent recalled second object behavior dataset can also reflect this characteristic, thereby increasing the amount of data in the second object behavior dataset, facilitating the recommendation of more information to users and improving the user experience.

[0108] S2073: Determine, based on a second preset database, a set of candidate behavior features whose similarity to the target behavior feature is greater than a second preset threshold; the second preset database stores the behavior features corresponding to the preset second object data;

[0109] In an embodiment of the present application, the second preset database can be updated regularly, and the update time can be set according to actual needs, for example, it can be set to update once a day. By regularly updating, the accuracy of the specific set of candidate behaviors can be improved, thereby improving the accuracy of the recommendation information.

[0110] S2075: Obtain a second object behavior data set based on the second object behavior data corresponding to each candidate behavior feature in the candidate behavior feature set.

[0111] In the embodiment of the present application, the second object behavior data set may include multiple second object behavior data.

[0112] In an embodiment of the present application, a second object behavior data set of interest migration can be obtained through the first object data and the second object data, thereby increasing the amount of second object data to be recommended, improving user experience, and increasing the exposure of recommended information.

[0113] S209: Determine the second object data to be recommended based on the candidate semantic second object data set and the candidate behavioral second object data set.

[0114] In the embodiments of this application, Figure 6 As shown, the determining of the second object data to be recommended based on the candidate semantic second object data set and the candidate behavioral second object data set includes:

[0115] S2091: Obtain the associated information type corresponding to each second object data in the candidate semantic second object data set and the candidate behavioral second object data set;

[0116] S2093: Determine the target associated information type according to the number of associated information corresponding to each associated information type;

[0117] S2095: Determine the number of first association information of the target association information type corresponding to each second object data, and the number of second association information corresponding to each second object data; the number of second association information is the sum of the number of association information corresponding to each second object data;

[0118] S2097: Determine the object data to be screened out based on the ratio of the number of first associated information to the number of second associated information corresponding to each second object data; the object data to be screened out is the second object data for which the ratio is less than a preset ratio threshold;

[0119] S2099: Deleting the object data to be screened out from the second candidate object dataset to obtain an updated second object dataset; the second candidate object dataset includes at least one of the second object semantic dataset and the second object behavior dataset;

[0120] In an embodiment of the present application, the set in which the object data to be screened out is located can be determined. The object data to be screened out can be data in the second object semantic data set or data in the second object behavior data set; if there are multiple object data to be screened out, the multiple object data to be screened out can be data in the second object semantic data set and data in the second object behavior data set.

[0121] S20911: Based on the updated second object data set, determine the second object data to be recommended.

[0122] In an embodiment of the present application, in order to ensure the effectiveness of each CP being recommended, a verticality threshold is applied to the CP to filter it, and CPs with high verticality in the category of publication are retained. For any second object data (CP), the categories of all articles corresponding to the CP can be obtained, and the proportion of the target category with the largest number of articles in the number of articles corresponding to all categories can be determined; and the second object data with a proportion less than a preset ratio threshold can be screened out; for example, a CP published ten articles, of which 8 were sports articles and 2 were entertainment articles, and the proportion of the category with the largest number of articles (sports) accounted for 80%. If the preset ratio threshold is set to 0.7, the CP can be determined as the second object data to be recommended. The higher the ratio threshold, the higher the professionalism of the second object data in a specific field, thereby ensuring the quality of the articles corresponding to the second object data and improving the click-through rate and exposure of the recommended information.

[0123] In the embodiments of this application, Figure 11 As shown, Figure 11 The associated information corresponding to the second target object data includes the title (title), the first-level category (category 1), the second-level category (category 2), and the tags (TAGS) to which the article belongs; Figure 12 is the association information corresponding to the semantic second object data to be recommended determined by the method of this embodiment; Figure 11-12 It can be seen that the titles (title), first-level categories (cate 1), second-level categories (cate 2) and article tags (TAGS) of the two articles have a high degree of similarity.

[0124] In the embodiments of this application, Figure 13 As shown, Figure 13Related information corresponding to another type of second target object data, including title (title), first-level category (category 1), second-level category (category 2), and article tags (TAGS); Figure 14 is the associated information corresponding to the second object data of the behavior to be recommended determined by the method of this embodiment; Figure 13-14 It can be seen that compared with Figure 13 , Figure 14 The recommended information is interest migration information, migrating from food to education; there is a certain behavioral divergence in categories, which is in line with the normal interest migration divergence of users; the number of recommended information is increased and the user experience is improved.

[0125] In the embodiments of this application, Figure 7 As shown, the determining of the second object data to be recommended based on the candidate semantic second object data set and the candidate behavioral second object data set includes:

[0126] S20901: Obtain comprehensive features of the second target object data;

[0127] In the embodiment of the present application, the corresponding comprehensive features can be determined based on the associated information of the second target object data.

[0128] S20903: Perform feature extraction on the second object semantic dataset and the second object behavior dataset to obtain comprehensive features of each object data in the second object dataset; the second object dataset includes the second object semantic dataset and the second object behavior dataset.

[0129] In an embodiment of the present application, the comprehensive features of each object data in the second object data set and the comprehensive features of the second target object data are of the same type of features; feature extraction can be performed on the above-mentioned second object semantic data set and the above-mentioned second object behavior data set based on the information recommendation model.

[0130] S20905: Calculate the similarity between each object data and the second target object data based on the comprehensive features of the second target object data and the comprehensive features of each object data in the second object data set.

[0131] S20907: Based on the similarity corresponding to each object data, obtain a ranking result of each object data in the second object data set.

[0132] In the embodiment of the present application, before obtaining the ranking result of each object data in the second object data set based on the similarity corresponding to each object data, the method further includes:

[0133] A first weight of the second object semantic data and a second weight of the second object behavioral data are determined.

[0134] In the embodiment of the present application, determining the first weight of the second object semantic data and the second weight of the second object behavior data includes:

[0135] Inputting the second sample object semantic data set, the second sample object behavior data set, the first preset weight, the second preset weight, and the sample consumption index of the sample object into the information recommendation model;

[0136] During the model training process, adjusting the first preset weight and the second preset weight until the consumption index output by the information recommendation model matches the sample consumption index of the sample object;

[0137] The first preset weight when matching is used as the first weight, and the second preset weight is used as the second weight.

[0138] In an embodiment of the present application, the consumption index may be used as a target to perform model training, thereby determining the weights of different second object data.

[0139] Accordingly, the ranking result of each object data in the second object data set is obtained based on the similarity corresponding to each object data, including:

[0140] Calculating the product of the similarity corresponding to each second object semantic data and the first weight to obtain a first updated similarity corresponding to each second object semantic data;

[0141] Calculating the product of the similarity corresponding to each second object behavior data and the second weight to obtain a second updated similarity corresponding to each second object behavior data;

[0142] According to the first update similarity corresponding to each second object semantic data and the second update similarity corresponding to each second object behavior data, each object data in the second object data set is sorted to obtain a sorting result for each object data in the second object data set.

[0143] In an embodiment of the present application, the similarities of different second object data can be updated according to the weights of the candidate semantic second object data and the candidate behavioral second object data, thereby improving the accuracy of sorting and the accuracy of recommendation information.

[0144] S20909: Determine the second object data to be recommended based on the ranking results of the respective object data in the second object data set.

[0145] In the embodiment of the present application, the object data in the second object data set may be sorted from large to small according to similarity, and a preset number of object data with the highest order may be used as the second object data to be recommended.

[0146] In an embodiment of the present application, a preset number of object data with the highest similarity to the second target object data can be determined from the second object data set, and the preset number of object data can be recommended to the target object; when the second object data is CP, a preset number of CPs to be recommended can be recommended to the target object.

[0147] S2011: Based on the second object data to be recommended, recommend associated information of the second object data to be recommended to the target object.

[0148] In an embodiment of the present application, the associated information of the second object data to be recommended includes both the semantic data of the second object and the behavioral data of the second object; the associated information may include, but is not limited to, text, pictures, videos, and other information. When the target object is a user, the first target object data and the second target object data can be obtained based on a single click of the user, thereby determining the second object data to be recommended; for example, when the second object data to be recommended is a publisher identifier to be recommended, the user can follow the publisher identifier to be recommended based on the recommendation, thereby facilitating the subsequent recommendation of the associated information of the publisher identifier to the user, the associated information including, but not limited to, text, pictures, videos, and other information published by the publisher identifier.

[0149] In a specific embodiment, Figure 15 As shown, on the article display page, the second object data corresponding to the article viewed by the user is the first second object data, and the user has already followed this second object data; based on this, the second second object data to be recommended, the second second object data, the third second object data, and so on can be determined; and each identifier and the corresponding "Follow" control are displayed on the current page. The user can click on any control to obtain the associated information corresponding to the second object data of interest (for example: article, video, etc.). This embodiment can ensure that the CPs to be recommended include both semantically similar CPs and CPs with reasonable interest migration.

[0150] It can be seen from the technical solutions provided by the above embodiments of the present application that the embodiments of the present application obtain the first target object data of the target object and the target association information corresponding to the above-mentioned first target object data; and determine the second target object data for publishing the above-mentioned target association information; based on the above-mentioned target association information, determine a second object semantic data set whose semantic similarity with the above-mentioned second target object data is greater than a first preset threshold; based on the above-mentioned first target object data and the above-mentioned second target object data, determine a second object behavior data set whose behavior similarity with the above-mentioned second target object data is greater than a second preset threshold; based on the above-mentioned second object semantic data set and the above-mentioned second object behavior data set, determine the second object data to be recommended; not only the second object semantic data set with semantic similarity to the second target object data is obtained, but also the second object behavior data set with interest migration is obtained, thereby increasing the number of second object data to be recommended; based on the above-mentioned second object data to be recommended, recommend the association information of the above-mentioned second object data to be recommended to the above-mentioned target object; realize the recall of the association information to be recommended from the second object data granularity, and increase the number of recommended information.

[0151] The present application also provides an information recommendation device, such as Figure 16 As shown, the device includes:

[0152] The target association information acquisition module 1610 is configured to acquire first target object data of a target object and target association information corresponding to the first target object data;

[0153] The second target object data determining module 1620 is used to determine the second target object data for publishing the target association information;

[0154] A second object semantic data set determining module 1630 is configured to determine, based on the target association information, a second object semantic data set having a semantic similarity with the second target object data greater than a first preset threshold;

[0155] A second object behavior dataset determining module 1640 is configured to determine, based on the first target object data and the second target object data, a second object behavior dataset having a behavior similarity with the second target object data that is greater than a second preset threshold;

[0156] A second object data to be recommended determining module 1650 is configured to determine the second object data to be recommended based on the second object semantic dataset and the second object behavior dataset;

[0157] The associated information determining module 1660 is configured to recommend the associated information of the second object data to be recommended to the target object based on the second object data to be recommended.

[0158] In some embodiments, the second object semantic dataset determining module may include:

[0159] a target semantic feature determination unit, configured to extract semantic features from the target association information to obtain target semantic features of the second target object data;

[0160] a candidate semantic feature set determining unit, configured to determine, based on a first preset database, a candidate semantic feature set having a similarity with the target semantic feature greater than the first preset threshold; wherein the first preset database stores semantic features corresponding to the second preset object data;

[0161] The second object semantic data set determining unit is configured to obtain the second object semantic data set based on the second object semantic data corresponding to each candidate semantic feature in the candidate semantic feature set.

[0162] In some embodiments, the second object behavior dataset determination module may include:

[0163] a target behavior feature determination unit, configured to extract behavior features from the first target object data and the second target object data based on a behavior feature extraction model to obtain target behavior features of the second target object data;

[0164] a candidate behavior feature set determining unit, configured to determine, based on a second preset database, a candidate behavior feature set having a similarity with the target behavior feature greater than a second preset threshold; the second preset database storing the behavior features corresponding to the second preset object data;

[0165] The second object behavior data set determining unit is configured to obtain the second object behavior data set based on the second object behavior data corresponding to each candidate behavior feature in the candidate behavior feature set.

[0166] In some embodiments, a behavior feature extraction model training module is further included, which includes:

[0167] A second sample object data set acquiring unit, configured to acquire a second sample object data set, wherein each second sample object data in the second sample object data set is annotated with a corresponding behavior feature label;

[0168] a second sample object data sequence construction unit, configured to use each second sample object data in the second sample object data set as a node, connect any two second sample object data that are clicked consecutively as an edge, and use the number of consecutive clicks on the any two second sample object data as the weight of the corresponding edge to construct the second sample object data sequence;

[0169] a behavior feature extraction training unit, configured to perform behavior feature extraction training on a preset machine learning model based on each second sample object data in the second sample object data sequence, so as to adjust model parameters of the preset machine learning model until a behavior feature label of each second sample object data output by the preset machine learning model matches the labeled behavior feature label;

[0170] The model determination unit is used to use the machine learning model corresponding to the model parameters when the output behavior feature label is matched with the marked behavior feature label as the behavior feature extraction model.

[0171] In some embodiments, the module for determining the second object data to be recommended may include:

[0172] an associated information type acquiring unit, configured to acquire an associated information type corresponding to each second object data in the second object semantic dataset and the second object behavior dataset;

[0173] a target association information type determining unit, configured to determine a target association information type according to the number of association information corresponding to each association information type;

[0174] a correlation information quantity determining unit, configured to determine the quantity of first correlation information of a target correlation information type corresponding to each second object data, and the quantity of second correlation information corresponding to each second object data;

[0175] a unit for determining object data to be screened out, configured to determine the object data to be screened out based on a ratio of the amount of first associated information to the amount of second associated information corresponding to each second object data; the object data to be screened out is second object data for which the ratio is less than a preset ratio threshold;

[0176] a second object dataset updating unit, configured to delete the object data to be screened out from the second candidate object dataset to obtain an updated second object dataset; the second candidate object dataset comprising at least one of the second object semantic dataset and the second object behavior dataset;

[0177] The second object data to be recommended determining unit is configured to determine the second object data to be recommended based on the updated second object data set.

[0178] In some embodiments, the module for determining the second object data to be recommended may include:

[0179] a comprehensive feature acquisition unit, configured to acquire comprehensive features of the second target object data;

[0180] a comprehensive feature extraction unit, configured to perform feature extraction on the second object semantic dataset and the second object behavior dataset to obtain comprehensive features of each object data in the second object dataset; the second object dataset includes the second object semantic dataset and the second object behavior dataset;

[0181] a similarity calculation unit, configured to calculate a similarity between each object data and the second target object data based on the comprehensive features of the second target object data and the comprehensive features of each object data in the second object data set;

[0182] a ranking result determining unit, configured to obtain a ranking result of each object data in the second object data set based on a similarity corresponding to each object data;

[0183] The to-be-recommended data determining unit is configured to determine the to-be-recommended second object data based on a ranking result of each object data in the second object data set.

[0184] In some embodiments, the above apparatus may further include:

[0185] The weight determination module is used to determine a first weight of the second object semantic data and a second weight of the second object behavior data.

[0186] In some embodiments, the ranking result determination unit may include:

[0187] a first updated similarity determination subunit, configured to calculate a product of a similarity corresponding to each second object semantic data and the first weight, to obtain a first updated similarity corresponding to each second object semantic data;

[0188] a second updated similarity determination subunit, configured to calculate a product of a similarity corresponding to each second object behavior data and the second weight to obtain a second updated similarity corresponding to each second object behavior data;

[0189] The sorting result determination subunit is used to sort the object data in the second object data set according to the first update similarity corresponding to each second object semantic data and the second update similarity corresponding to each second object behavior data, to obtain the sorting result of each object data in the second object data set.

[0190] In some embodiments, the target association information acquisition module may include:

[0191] a candidate association information acquisition unit, configured to acquire a plurality of first object data of the target object and a plurality of candidate association information corresponding to the plurality of first object data;

[0192] A display duration acquisition unit, used to acquire the display duration of each candidate association information;

[0193] a target association information determination unit, configured to determine candidate association information having a display duration greater than a preset duration threshold as the target association information;

[0194] The first target object data determining unit is configured to determine the first object data corresponding to the target association information as the first target object data.

[0195] The device and method embodiments in the device embodiments are based on the same inventive concept.

[0196] An embodiment of the present application provides an information recommendation device, which includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the information recommendation method provided in the above method embodiment.

[0197] An embodiment of the present application also provides a computer storage medium, which can be set in a terminal to store at least one instruction or at least one program related to implementing an information recommendation method in a method embodiment. The at least one instruction or at least one program is loaded and executed by the processor to implement the information recommendation method provided by the above method embodiment.

[0198] Embodiments of the present application also provide a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to implement the information recommendation method provided in the above method embodiment.

[0199] Optionally, in an embodiment of the present application, the storage medium may be located in at least one of a plurality of network servers in a computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk, or an optical disk.

[0200] The memory described in the embodiment of the present application can be used to store software programs and modules, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory. The memory may mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, application programs required for functions, etc.; the data storage area can store data created according to the use of the device, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device. Accordingly, the memory may also include a memory controller to provide the processor with access to the memory.

[0201] The information recommendation method provided in the embodiment of the present application can be executed in a mobile terminal, a computer terminal, a server or a similar computing device. Taking running on a server as an example, Figure 17 This is a hardware structure diagram of a server of an information recommendation method provided by an embodiment of the present application. Figure 17 As shown, the server 1700 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPUs) 1710 (the processor 1710 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 1730 for storing data, and one or more storage media 1720 (such as one or more mass storage devices) for storing application programs 1723 or data 1722. Among them, the memory 1730 and the storage medium 1720 can be temporary storage or permanent storage. The program stored in the storage medium 1720 may include one or more modules, each module may include a series of instruction operations on the server. Furthermore, the central processing unit 1710 can be configured to communicate with the storage medium 1720 to execute a series of instruction operations in the storage medium 1720 on the server 1700. The server 1700 may also include one or more power supplies 1760, one or more wired or wireless network interfaces 1750, one or more input and output interfaces 1740, and / or one or more operating systems 1721, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0202] The input / output interface 1740 can be used to receive or send data via a network. Specific examples of the aforementioned network may include a wireless network provided by the communication provider of the server 1700. In one embodiment, the input / output interface 1740 includes a network interface controller (NIC), which can be connected to other network devices via a base station to communicate with the Internet. In another embodiment, the input / output interface 1740 can be a radio frequency (RF) module for wirelessly communicating with the Internet.

[0203] It can be understood by those skilled in the art that Figure 17 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 17 More or fewer components than shown, or with Figure 17 Different configurations shown.

[0204] It can be seen from the embodiments of the information recommendation method, device, equipment or storage medium provided by the above-mentioned present application that the present application obtains the first target object data of the target object and the target association information corresponding to the first target object data; and determines the second target object data for publishing the target association information; based on the target association information, determines a second object semantic data set whose semantic similarity with the second target object data is greater than a first preset threshold; based on the first target object data and the second target object data, determines a second object behavior data set whose behavior similarity with the second target object data is greater than a second preset threshold; based on the second object semantic data set and the second object behavior data set, determines the second object data to be recommended; not only the second object semantic data set that is semantically similar to the second target object data is obtained, but also the second object behavior data set of interest migration is obtained, thereby increasing the number of second object data to be recommended; based on the second object data to be recommended, recommends the association information of the second object data to be recommended to the target object; realizes the recall of the association information to be recommended from the second object data granularity, and increases the number of recommended information.

[0205] It should be noted that the order of the embodiments of the present application described above is for descriptive purposes only and does not represent the superiority or inferiority of the embodiments. The above description is of specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0206] The various embodiments in this specification are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from the other embodiments. In particular, the device, equipment, and storage medium embodiments are generally similar to the method embodiments, so their descriptions are relatively simplified. For relevant portions, refer to the descriptions of the method embodiments.

[0207] Those skilled in the art will understand that all or part of the steps of implementing the above embodiments can be completed by hardware, or can be completed by a program to instruct the relevant hardware. The program can be stored in a computer storage medium, and the above-mentioned storage medium can be a read-only memory, a disk or an optical disk, etc.

[0208] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. An information recommendation method, characterized in that: The method comprises: Acquire first target object data of a target object and target association information corresponding to the first target object data; Determine the second target object data for publishing the target association information; Determining, based on the target association information, a second object semantic data set having a semantic similarity with the second target object data greater than a first preset threshold; Based on a behavior feature extraction model, extracting behavior features of the first target object data and the second target object data to obtain target behavior features of the second target object data; Determining, based on a second preset database, a set of candidate behavioral features whose similarity to the target behavioral features is greater than a second preset threshold; the second preset database stores behavioral features corresponding to the second preset object data; obtaining a second object behavior data set based on the second object behavior data corresponding to each candidate behavior feature in the candidate behavior feature set; determining second object data to be recommended based on the second object semantic dataset and the second object behavior dataset; recommending associated information of the second object data to be recommended to the target object based on the second object data to be recommended; The training method of the behavior feature extraction model includes: Acquire a second sample object data set, wherein each second sample object data in the second sample object data set is marked with a corresponding behavior feature label; constructing a second sample object data sequence by taking each second sample object data in the second sample object data set as a node, connecting any two second sample object data that are clicked consecutively as an edge, and taking the number of times the two second sample object data are clicked consecutively as the weight of the corresponding edge; Based on each second sample object data in the second sample object data sequence, performing behavior feature extraction training on a preset machine learning model to adjust model parameters of the preset machine learning model until a behavior feature label of each second sample object data output by the preset machine learning model matches the annotated behavior feature label; The machine learning model corresponding to the model parameters when the output behavior feature label matches the annotated behavior feature label is used as the behavior feature extraction model.

2. The method according to claim 1, characterized in that The determining, based on the target association information, a second object semantic data set having a semantic similarity with the second target object data greater than a first preset threshold, includes: performing semantic feature extraction on the target association information to obtain target semantic features of the second target object data; Determining, based on a first preset database, a set of candidate semantic features whose similarity to the target semantic feature is greater than a first preset threshold; the first preset database stores semantic features corresponding to second preset object data; A second object semantic data set is obtained based on the second object semantic data corresponding to each candidate semantic feature in the candidate semantic feature set.

3. The method according to any one of claims 1 to 2, characterized in that The determining the second object data to be recommended based on the second object semantic dataset and the second object behavior dataset includes: Acquire the associated information type corresponding to each second object data in the second object semantic dataset and the second object behavior dataset; Determine the target associated information type according to the number of associated information corresponding to each associated information type; Determine the number of first association information of the target association information type corresponding to each second object data, and the number of second association information corresponding to each second object data; the second association information number is the sum of the numbers of association information corresponding to each second object data; Determining the object data to be screened out according to the ratio of the amount of the first associated information to the amount of the second associated information corresponding to each second object data; the object data to be screened out is the second object data for which the ratio is less than a preset ratio threshold; Deleting the object data to be screened out from the second candidate object dataset to obtain an updated second object dataset; the second candidate object dataset includes at least one of the second object semantic dataset and the second object behavior dataset; The second object data to be recommended is determined based on the updated second object data set.

4. The method according to any one of claims 1 to 2, characterized in that The determining the second object data to be recommended based on the second object semantic dataset and the second object behavior dataset includes: Acquiring comprehensive features of the second target object data; Performing feature extraction on the second object semantic dataset and the second object behavior dataset to obtain comprehensive features of each object data in the second object dataset; the second object dataset includes the second object semantic dataset and the second object behavior dataset; calculating a similarity between each object data and the second target object data based on the comprehensive features of the second target object data and the comprehensive features of each object data in the second object data set; Obtaining a ranking result of each object data in the second object data set based on the similarity corresponding to each object data; The second object data to be recommended is determined based on the ranking result of each object data in the second object data set.

5. The method according to claim 4, characterized in that Before obtaining the ranking result of each object data in the second object data set based on the similarity corresponding to each object data, the method further includes: determining a first weight of the second object semantic data and a second weight of the second object behavioral data; Accordingly, obtaining the ranking result of each object data in the second object data set based on the similarity corresponding to each object data includes: Calculating the product of the similarity corresponding to each second object semantic data and the first weight to obtain a first updated similarity corresponding to each second object semantic data; Calculating the product of the similarity corresponding to each second object behavior data and the second weight to obtain a second updated similarity corresponding to each second object behavior data; The object data in the second object data set are sorted according to the first update similarity corresponding to each second object semantic data and the second update similarity corresponding to each second object behavior data to obtain a sorting result for each object data in the second object data set.

6. The method according to any one of claims 1 to 2, characterized in that The acquiring first target object data of the target object and target association information corresponding to the first target object data includes: Acquire multiple first object data of the target object and multiple candidate association information corresponding to the multiple first object data; Get the display duration of each candidate association information; Determine the candidate association information whose display duration is greater than a preset duration threshold as the target association information; The first object data corresponding to the target association information is determined as the first target object data.

7. An information recommendation device, characterized in that: The device comprises: a target association information acquisition module, configured to acquire first target object data of a target object and target association information corresponding to the first target object data; A second target object data determining module, configured to determine the second target object data for publishing the target association information; A second object semantic data set determining module, configured to determine, based on the target association information, a second object semantic data set having a semantic similarity with the second target object data greater than a first preset threshold; a second object behavior data set determination module, configured to determine, based on the first target object data and the second target object data, a second object behavior data set whose behavior similarity with the second target object data is greater than a second preset threshold; the second object behavior data set determination module comprising: a target behavior feature determination unit, configured to extract behavior features from the first target object data and the second target object data based on a behavior feature extraction model to obtain target behavior features of the second target object data; a candidate behavior feature set determination unit, configured to determine, based on a second preset database, a candidate behavior feature set whose similarity with the target behavior features is greater than the second preset threshold; the second preset database stores behavior features corresponding to the second preset object data; a second object behavior data set determination unit, configured to obtain a second object behavior data set based on the second object behavior data corresponding to each candidate behavior feature in the candidate behavior feature set; a second object data to be recommended determining module, configured to determine the second object data to be recommended based on the second object semantic data set and the second object behavior data set; an associated information determining module, configured to recommend associated information of the second object data to be recommended to the target object based on the second object data to be recommended; A second sample object data set acquiring unit, configured to acquire a second sample object data set, wherein each second sample object data in the second sample object data set is annotated with a corresponding behavior feature label; a second sample object data sequence construction unit, configured to use each second sample object data in the second sample object data set as a node, connect any two second sample object data that are clicked consecutively as an edge, and use the number of consecutive clicks on the any two second sample object data as the weight of the corresponding edge to construct the second sample object data sequence; a behavior feature extraction training unit, configured to perform behavior feature extraction training on a preset machine learning model based on each second sample object data in the second sample object data sequence, so as to adjust model parameters of the preset machine learning model until a behavior feature label of each second sample object data output by the preset machine learning model matches the labeled behavior feature label; The model determination unit is used to use the machine learning model corresponding to the model parameters when the output behavior feature label is matched with the marked behavior feature label as the behavior feature extraction model.

8. A computer storage medium, characterized in that The computer storage medium stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the information recommendation method according to any one of claims 1 to 6.

Images