Information recommendation method and device, electronic equipment and computer readable storage medium

Abstract

The embodiment of the invention provides an information recommendation method and device, electronic equipment and a storage medium, relates to the technical field of machine learning, and can be applied to scenes such as artificial intelligence, intelligent traffic and auxiliary driving. The method comprises the steps of obtaining object features of a recommended object and recommendation features of at least two pieces of information to be recommended; for each recommendation feature, feature coding is carried out based on the similarity between the object feature and the recommendation feature, and coding information corresponding to the similarity is obtained; determining position information of each numerical value in the coding information, and generating fusion features of the object features and the recommendation features based on the position information and the coding information; inputting the fusion features into a pre-trained information recommendation model to obtain a matching degree between the recommended object and each piece of to-be-recommended information; and screening target recommendation information from the to-be-recommended information based on the matching degree, and recommending the target recommendation information to the recommended object. According to the embodiment of the invention, the accuracy of information recommendation is improved through coding and fusion operation for the feature information.

Description

Technical Field

[0001] The present application relates to the technical field of machine learning. Specifically, the present application relates to an information recommendation method, device, electronic device and computer-readable storage medium. Background Art

[0002] With the rapid growth of the Internet and information resources, the problem of information overload has also arisen, and it is difficult for users to obtain the information they need. Information recommendation systems with information push as the service model have emerged. Information recommendation refers to the system recommending useful information that is suitable but cannot be quickly obtained to the recommended object, so that the recommended object can efficiently obtain the required information.

[0003] The existing information recommendation system can train the information recommendation model based on machine learning, and recommend information to the recommended object based on the pre-trained information recommendation model; in the training and prediction process of the information recommendation model, due to the different feature dimensions of the recommended object and the information to be recommended, the features of the recommended object and the information to be recommended can be discretized to obtain fused features, and the fused features are input into the information recommendation model, but the information recommendation model is often unable to effectively adapt to the above fused features, so that the information recommendation model cannot fully learn the information preferences of the recommended object, resulting in the accuracy and pertinence of the recommended information being low. Summary of the invention

[0004] The embodiments of the present application provide an information recommendation method, device, electronic device, and computer-readable storage medium, which can solve the problem of low accuracy and pertinence of information recommendation. The technical solution is as follows:

[0005] According to one aspect of an embodiment of the present application, there is provided an information recommendation method, the method comprising:

[0006] Obtaining object features of a recommended object and at least two recommendation features of information to be recommended;

[0007] For each recommended feature, determine the similarity between the object feature and the recommended feature; perform feature encoding based on the similarity between the object feature and the recommended feature to obtain encoding information corresponding to the similarity; determine the position information of each value in the encoding information, and generate a fusion feature of the object feature and the recommended feature based on the position information of each value and the encoding information;

[0008] Input each fusion feature into the pre-trained information recommendation model for data prediction to obtain the matching degree between the recommended object and each information to be recommended;

[0009] The target recommended information is filtered out from the information to be recommended based on the matching degree, and the target recommended information is recommended to the recommended object.

[0010] In a possible implementation, the above-mentioned encoding information includes binary encoding;

[0011] Based on the location information and encoding information of each value, a fusion feature of the object feature and the recommendation feature is generated, including:

[0012] The position information of each bit value of the binary code is converted into a vector to obtain the position vector corresponding to each bit value, and the position vectors are fused to obtain the fused feature.

[0013] In a possible implementation, the feature encoding is performed based on the similarity between the object feature and the recommended feature to obtain the encoding information corresponding to the similarity, including:

[0014] Adjust the value of the similarity based on a preset numerical range to obtain a target similarity;

[0015] Perform binary conversion on the target similarity and determine the binary code corresponding to the similarity.

[0016] In another possible implementation, the position information of each bit value of the binary code is converted into a vector to obtain a position vector corresponding to each bit value, and the position vectors are fused to obtain a fused feature, including:

[0017] For each bit of the numerical value, the position vector corresponding to the numerical value is searched in a preset vector conversion table; wherein the vector table includes the correlation between the position information of the numerical value and the position vector;

[0018] Based on the dimension of the position vector, each position vector is added bit by bit to obtain the fusion feature of the object feature and the recommendation feature.

[0019] In another possible implementation, the method further includes:

[0020] Obtaining object information of the recommended object, numerically encoding the object information, and obtaining candidate object features of the recommended object;

[0021] For each piece of information to be recommended, numerically encode the information to be recommended to obtain candidate recommendation features of the information to be recommended; concatenate the candidate recommendation features with the candidate object features to obtain concatenated features corresponding to the information to be recommended;

[0022] The above-mentioned input of each fusion feature into the pre-trained information recommendation model for data prediction, obtains the matching degree between the recommended object and each information to be recommended, including:

[0023] For each piece of information to be recommended, the corresponding fusion features and concatenation features are input into the pre-trained information recommendation model for data prediction to obtain the matching degree between the recommended object and the information to be recommended.

[0024] In another possible implementation, the corresponding fusion features and concatenation features are input into a pre-trained information recommendation model for data prediction to obtain the matching degree between the recommended object and the information to be recommended, including:

[0025] The splicing feature and the fusion feature are spliced ​​to obtain the fusion splicing feature;

[0026] Based on the information recommendation model, data prediction is performed on the fused splicing features to obtain the matching degree between the recommended object and the information to be recommended.

[0027] In another possible implementation, the above information recommendation model is trained based on the following method:

[0028] Obtaining sample object features of a historically recommended object, and obtaining sample recommendation features of a plurality of historically recommended information corresponding to the historically recommended object;

[0029] Obtain access records of historical recommendation information, and generate label matching degrees between each historical recommendation information and the historical recommendation object based on the access records;

[0030] For each sample recommendation feature, determine the sample similarity between the sample object feature and the sample recommendation feature; perform feature encoding based on the sample similarity between the sample object feature and the sample recommendation feature to obtain sample encoding information corresponding to the sample similarity; determine the sample position information of each value in the sample encoding information, and generate a sample fusion feature of the sample object feature and the sample recommendation feature based on the sample position information of each value and the sample encoding information;

[0031] Input each sample fusion feature into the preset initial model for data prediction to obtain the sample matching degree between each historical recommendation information and the historical recommendation object;

[0032] Based on the matching degree of each sample and the matching degree of the corresponding label, the initial model is updated at least once until the preset update end condition is met, and the initial model after the last update is used as the information recommendation model;

[0033] The update operation includes:

[0034] Calculate the difference between the sample matching degree and the corresponding label matching degree according to the preset loss function;

[0035] Update the parameters of the initial model based on the difference;

[0036] The update end condition includes: the difference is not greater than a preset convergence threshold.

[0037] According to another aspect of an embodiment of the present application, there is provided an information recommendation device, the device comprising:

[0038] An acquisition module, used to acquire object features of a recommended object and at least two recommendation features of information to be recommended;

[0039] A fusion module is used to determine the similarity between the object feature and the recommended feature for each recommended feature; perform feature encoding based on the similarity between the object feature and the recommended feature to obtain encoding information corresponding to the similarity; determine the position information of each value in the encoding information, and generate a fusion feature of the object feature and the recommended feature based on the position information of each value and the encoding information;

[0040] A prediction module is used to input each fusion feature into a pre-trained information recommendation model for data prediction to obtain the matching degree between the recommended object and each information to be recommended;

[0041] The recommendation module is used to filter out target recommendation information from the information to be recommended based on the matching degree, and recommend the target recommendation information to the recommended object.

[0042] In a possible implementation, the above-mentioned encoding information includes binary encoding;

[0043] When the fusion module generates the fusion feature of the object feature and the recommendation feature based on the position information and the encoding information of each value, it is used to:

[0044] The position information of each bit value of the binary code is converted into a vector to obtain the position vector corresponding to each bit value, and the position vectors are fused to obtain the fused feature.

[0045] In a possible implementation, when the fusion module performs feature encoding based on the similarity between the object feature and the recommended feature to obtain encoding information corresponding to the similarity, it is used to:

[0046] Adjust the value of the similarity based on a preset numerical range to obtain a target similarity;

[0047] Perform binary conversion on the target similarity and determine the binary code corresponding to the similarity.

[0048] In another possible implementation, the fusion module performs vector conversion on the position information of each bit value of the binary code to obtain the position vector corresponding to each bit value, and fuses the position vectors to obtain the fusion feature, which is used to:

[0049] For each bit of the numerical value, the position vector corresponding to the numerical value is searched in a preset vector conversion table; wherein the vector table includes the correlation between the position information of the numerical value and the position vector;

[0050] Based on the dimension of the position vector, each position vector is added bit by bit to obtain the fusion feature of the object feature and the recommendation feature.

[0051] In yet another possible implementation, the acquisition module is further configured to:

[0052] Obtaining object information of the recommended object, numerically encoding the object information, and obtaining candidate object features of the recommended object;

[0053] For each piece of information to be recommended, numerically encode the information to be recommended to obtain candidate recommendation features of the information to be recommended; concatenate the candidate recommendation features with the candidate object features to obtain concatenated features corresponding to the information to be recommended;

[0054] The above prediction module is used to input each fusion feature into the pre-trained information recommendation model for data prediction to obtain the matching degree between the recommended object and each information to be recommended:

[0055] For each piece of information to be recommended, the corresponding fusion features and concatenation features are input into the pre-trained information recommendation model for data prediction to obtain the matching degree between the recommended object and the information to be recommended.

[0056] In another possible implementation, when the prediction module inputs the corresponding fusion features and concatenation features into the pre-trained information recommendation model for data prediction and obtains the matching degree between the recommended object and the information to be recommended, it is used to:

[0057] The splicing feature and the fusion feature are spliced ​​to obtain the fusion splicing feature;

[0058] Based on the information recommendation model, data prediction is performed on the fused splicing features to obtain the matching degree between the recommended object and the information to be recommended.

[0059] In another possible implementation, the above information recommendation model is trained based on the following method:

[0060] Obtaining sample object features of a historically recommended object, and obtaining sample recommendation features of a plurality of historically recommended information corresponding to the historically recommended object;

[0061] Obtain access records of historical recommendation information, and generate label matching degrees between each historical recommendation information and the historical recommendation object based on the access records;

[0062] For each sample recommendation feature, determine the sample similarity between the sample object feature and the sample recommendation feature; perform feature encoding based on the sample similarity between the sample object feature and the sample recommendation feature to obtain sample encoding information corresponding to the sample similarity; determine the sample position information of each value in the sample encoding information, and generate a sample fusion feature of the sample object feature and the sample recommendation feature based on the sample position information of each value and the sample encoding information;

[0063] Input each sample fusion feature into the preset initial model for data prediction to obtain the sample matching degree between each historical recommendation information and the historical recommendation object;

[0064] Based on the matching degree of each sample and the matching degree of the corresponding label, the initial model is updated at least once until the preset update end condition is met, and the initial model after the last update is used as the information recommendation model;

[0065] The update operation includes:

[0066] Calculate the difference between the sample matching degree and the corresponding label matching degree according to the preset loss function;

[0067] Update the parameters of the initial model based on the difference;

[0068] The update end condition includes: the difference is not greater than a preset convergence threshold.

[0069] According to another aspect of an embodiment of the present application, an electronic device is provided, which includes: a memory, a processor, and a computer program stored in the memory, and the processor executes the computer program to implement the steps of the method shown in the first aspect of the embodiment of the present application.

[0070] According to another aspect of the embodiments of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method shown in the first aspect of the embodiments of the present application are implemented.

[0071] According to one aspect of an embodiment of the present application, a computer program product is provided, which includes a computer program, and when the computer program is executed by a processor, the steps of the method shown in the first aspect of the embodiment of the present application are implemented.

[0072] The beneficial effects of the technical solution provided by the embodiment of the present application are:

[0073] The embodiment of the present application obtains the object feature of the recommended object and the recommendation features of at least two pieces of information to be recommended, determines the similarity between the object feature and each recommendation feature, generates a fusion feature of the object feature and each recommendation feature based on the encoding information corresponding to the similarity, and the position information of each numerical value in the encoding information; then inputs each fusion feature into the pre-trained information recommendation model for data prediction, and then obtains the matching degree between the recommended object and each piece of information to be recommended, so as to filter out the target recommendation information from the information to be recommended based on the matching degree, and recommend the target recommendation information to the recommended object. The embodiment of the present application can encode the similarity between the object feature and each recommendation feature, obtain the encoding information corresponding to the similarity, and then perform feature fusion according to the encoding information and the position information of each numerical value in the encoding information, and generate a fusion feature of the object feature and the recommendation feature, the fusion feature can effectively characterize the association between the recommended object and the information to be recommended, and can be well adapted to the information recommendation model, thereby improving the accuracy of the matching degree between the recommended object and the information to be recommended.

[0074] Different from the prior art in which the features of the recommended object and the information to be recommended are discretized to obtain fused features, the feature fusion through similarity encoding in this application can retain the complete original feature information, enable the information recommendation model to fully learn the information preferences of the recommended object, and ensure the accuracy and pertinence of the target recommended information. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in describing the embodiments of the present application.

[0076] Figure 1 A schematic diagram of an application scenario of an information recommendation method provided in an embodiment of the present application;

[0077] Figure 2 A flowchart of an information recommendation method provided in an embodiment of the present application;

[0078] Figure 3 A schematic diagram of a process of feature fusion in an information recommendation method provided in an embodiment of the present application;

[0079] Figure 4 A schematic diagram of a process of vector conversion in an information recommendation method provided in an embodiment of the present application;

[0080] Figure 5 A schematic diagram of a process for determining a matching degree in an information recommendation method provided in an embodiment of the present application;

[0081] Figure 6 A schematic diagram of another process for determining a matching degree in an information recommendation method provided in an embodiment of the present application;

[0082] Figure 7 A schematic diagram of a process of model training in an information recommendation method provided in an embodiment of the present application;

[0083] Figure 8 A flowchart of an exemplary information recommendation method provided in an embodiment of the present application;

[0084] Fig. 9 A schematic diagram of the structure of an information recommendation device provided in an embodiment of the present application;

[0085] Fig.10 A schematic diagram of the structure of an information recommendation electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0086] The embodiments of the present application are described below in conjunction with the drawings in the present application. It should be understood that the implementation methods described below in conjunction with the drawings are exemplary descriptions for explaining the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions of the embodiments of the present application.

[0087] It will be understood by those skilled in the art that, unless specifically stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the terms "including" and "comprising" used in the embodiments of the present application refer to that the corresponding features can be implemented as the presented features, information, data, steps, operations, elements and / or components, but do not exclude the implementation as other features, information, data, steps, operations, elements, components and / or combinations thereof supported by the technical field. It should be understood that when we say that an element is "connected" or "coupled" to another element, the one element may be directly connected or coupled to the other element, or it may refer to that the one element and the other element establish a connection relationship through an intermediate element. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein indicates at least one of the items defined by the term, for example, "A and / or B" may be implemented as "A", or as "B", or as "A and B".

[0088] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0089] In recent years, the development and popularization of the Internet has provided people with a new carrier for information storage, processing, transmission and use. Network information has quickly become one of the main channels for social members to obtain knowledge and information. As a human-computer interaction system, information recommendation systems have been widely used in all aspects of social life. Therefore, it is of great significance to improve the pertinence and accuracy of information recommendation systems.

[0090] In the prior art, the features of the recommended object and the information to be recommended can usually be used as a sparse matrix, and each floating point number of the sparse matrix is ​​discretized and bucketed to obtain a fusion matrix, which is input into the information recommendation model for model training or prediction; however, this method ignores the relative relationship between each bit value of the feature, which will lead to the loss of feature information. In other methods, the features of the recommended object and the information to be recommended can also be transformed into a fusion matrix to better retain the complete original feature information, but since the fusion matrix cannot adapt to the data distribution of the current information recommendation model, it is difficult to ensure the good learning and prediction effect of the information recommendation model in actual practice.

[0091] The information recommendation method, device, electronic device and computer-readable storage medium provided in this application are intended to solve the above technical problems in the prior art.

[0092] The following describes several exemplary embodiments to illustrate the technical solutions of the embodiments of the present application and the technical effects produced by the technical solutions of the present application. It should be noted that the following embodiments can refer to, draw on or combine with each other, and the same terms, similar features and similar implementation steps in different embodiments will not be described repeatedly.

[0093] like Figure 1 As shown, the information recommendation method of the present application can be applied to Figure 1 In the scenario shown, specifically, the server obtains the object features of the recommended object and the recommendation features of at least two pieces of information to be recommended, and for each recommendation feature, determines the similarity between the object feature and the recommendation feature; then, based on the similarity between the object feature and the recommendation feature, feature encoding is performed to obtain encoding information corresponding to the similarity, and based on the encoding information and the position information of each numerical value in the encoding information, a fusion feature of the object feature and the recommendation feature is generated; then, each fusion feature is input into a pre-trained information recommendation model for data prediction to obtain the matching degree between the recommended object and each piece of information to be recommended; based on the matching degree, the target recommendation information is screened out from the information to be recommended, and the target recommendation information is sent to the client, thereby realizing information recommendation for the object to be recommended.

[0094] Figure 1 In the scenario shown, the above information recommendation method can be performed in a server, and in other scenarios, it can also be performed in a terminal.

[0095] Those skilled in the art can understand that the "terminal" used here includes but is not limited to mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle terminals, aircraft, etc.; "server" can be implemented with an independent server or a server cluster composed of multiple servers. The embodiments of the present invention can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, assisted driving, etc.

[0096] The present application embodiment provides an information recommendation method, such as Figure 2 As shown, it can be applied to a server or terminal for information recommendation, and the method includes:

[0097] S201, obtaining object features of a recommended object and at least two recommendation features of information to be recommended.

[0098] Specifically, a server or terminal used for information recommendation can determine the object information of the recommended object and the corresponding at least two pieces of information to be recommended; then, based on the word vector model, vector mapping is performed on the above object information and each piece of information to be recommended to obtain the object features of the recommended object and the recommendation features of at least two pieces of information to be recommended.

[0099] Among them, the object features and recommendation features can be Embedding word vectors.

[0100] In an embodiment of the present application, the Embedding vector can be obtained by mapping based on the Word2vec (a word embedding model used to generate word vectors) model. Word embedding is a general term for language models and representation learning techniques in natural language processing (NLP). Conceptually, it refers to embedding a high-dimensional space with a dimension equal to the number of all words into a continuous vector space with a much lower dimension, and each word or phrase is mapped to a vector on the real number field. Word embedding methods include artificial neural networks, dimensionality reduction of word co-occurrence matrices, probability models, and explicit representations of the context in which the words are located. Word vectors have good semantic properties and are a common way to represent word features. The value of each dimension of a word vector represents a feature with a certain semantic and grammatical interpretation. Therefore, each dimension of a word vector can be called a word feature.

[0101] In some embodiments, taking the case where the recommended information is an advertisement, the recommended object may be a user or an advertisement spot in a web page; when the recommended object is a user, the object information may include the user's gender, age, job, etc.; when the recommended object is an advertisement spot in a web page, the object information may include contextual information in the web page, the theme of the web page, etc.

[0102] S202, for each recommended feature, determine the similarity between the object feature and the recommended feature; perform feature encoding based on the similarity between the object feature and the recommended feature to obtain encoding information corresponding to the similarity; determine the position information of each numerical value in the encoding information, and generate a fusion feature of the object feature and the recommended feature based on the position information of each numerical value and the encoding information.

[0103] The object feature can be represented by a first vector, and the recommendation feature can be represented by a second vector; the similarity can be determined based on the angle between the first vector and the second vector:

[0104] In some implementations, the inner product of the first vector and the second vector may be used as the similarity between the object feature and the recommended feature;

[0105] In other implementations, the cosine value of the angle between the first vector and the second vector may be used as the similarity between the object feature and the recommended feature, which is not specifically limited in the embodiments of the present application.

[0106] Optionally, the feature encoding may be a binary encoding; specifically, a server or terminal for information recommendation may scale the similarity between the object feature and the recommended feature based on a preset value range to obtain a target similarity, and then use the binary encoding of the target similarity as the corresponding encoding information.

[0107] S203, input each fusion feature into a pre-trained information recommendation model for data prediction, and obtain the matching degree between the recommended object and each information to be recommended.

[0108] The above-mentioned matching degree may include the click rate, conversion rate or value of the information to be recommended.

[0109] In some embodiments, when the recommended object is a user and the information to be recommended is an advertisement, the corresponding matching degree may be the click rate or forwarding rate of the user relative to the advertisement, and the matching degree may represent the user's interest in the advertisement.

[0110] In other embodiments, when the recommended object is an advertising spot in a public account or application and the information to be recommended is an advertisement, the corresponding matching degree may be the conversion rate of the advertising spot for the displayed advertisement, that is, the account registration rate, download rate, attention rate or form submission rate of the advertisement relative to the advertising spot.

[0111] In some other implementations, when the recommended object is an advertiser and the information to be recommended is an advertisement, the corresponding matching degree may be the bid of the advertiser for the advertisement.

[0112] Optionally, the information recommendation model can be an artificial neural network model. Artificial neural network is a non-programmed, adaptive, brain-style information processing. Its essence is to obtain a parallel distributed information processing function through the transformation and dynamic behavior of the network, and imitate the information processing function of the human brain nervous system to varying degrees and levels. It is an interdisciplinary subject involving multiple fields such as neuroscience, thinking science, artificial intelligence, and computer science.

[0113] Artificial neural networks include processing units such as input units, output units, and hidden units. The input unit is used to receive input information, namely fusion features; the output unit is used to output the system processing results, namely the matching degree between the recommended object and the information to be recommended; the hidden unit is located between the input and output units and is used to predict the data of the input fusion features. The connection weights between neurons reflect the connection strength between units, and the representation and processing of information are reflected in the connection relationship of the network processing units.

[0114] S204, filtering out target recommended information from the information to be recommended based on the matching degree, and recommending the target recommended information to the recommended object.

[0115] Specifically, the server or terminal for information recommendation can filter out target recommended information from the information to be recommended based on the matching degree and a preset screening threshold, and send the target recommended information to the client so that the recommended object obtains the target recommended information.

[0116] In the embodiment of the present application, taking the case where the information to be recommended is an advertisement and the recommended object is an advertisement position in a web page or a public account as an example, the corresponding matching degree may include the account attention rate, APP download rate, etc. obtained by the advertisement in its advertisement position. The account attention rate and the APP download volume may be weighted and summed based on the preset weights to obtain the final matching degree, and the advertisement corresponding to the final matching degree greater than the preset threshold may be used as the target advertisement, and the target advertisement may be pushed to the advertisement position for display, so as to achieve targeted push based on the advertisement position and improve the conversion rate of the advertisement position.

[0117] The embodiment of the present application obtains the object feature of the recommended object and the recommendation features of at least two pieces of information to be recommended, determines the similarity between the object feature and each recommendation feature, generates a fusion feature of the object feature and each recommendation feature based on the encoding information corresponding to the similarity, and the position information of each numerical value in the encoding information; then inputs each fusion feature into the pre-trained information recommendation model for data prediction, and then obtains the matching degree between the recommended object and each piece of information to be recommended, so as to filter out the target recommendation information from the information to be recommended based on the matching degree, and recommend the target recommendation information to the recommended object. The embodiment of the present application can encode the similarity between the object feature and each recommendation feature, obtain the encoding information corresponding to the similarity, and then perform feature fusion according to the encoding information and the position information of each numerical value in the encoding information, and generate a fusion feature of the object feature and the recommendation feature, the fusion feature can effectively characterize the association between the recommended object and the information to be recommended, and can be well adapted to the information recommendation model, thereby improving the accuracy of the matching degree between the recommended object and the information to be recommended.

[0118] Different from the prior art in which the features of the recommended object and the information to be recommended are discretized to obtain fused features, the feature fusion through similarity encoding in this application can retain the complete original feature information, enable the information recommendation model to fully learn the information preferences of the recommended object, and ensure the accuracy and pertinence of the target recommended information.

[0119] A possible implementation method is provided in an embodiment of the present application, and the above-mentioned encoding information includes binary encoding.

[0120] The above-mentioned fusion features of the object features and the recommended features are generated based on the position information and encoding information of each value, including:

[0121] The position information of each bit value of the binary code is converted into a vector to obtain the position vector corresponding to each bit value, and the position vectors are fused to obtain the fused feature.

[0122] Specifically, a binary vector conversion table may be pre-constructed, and the vector conversion table may be used to query the position vector of each bit value in the binary code.

[0123] In some implementations, the maximum value K of the length of the binary code corresponding to each recommended feature may be determined first, and then K position vectors with a dimension of M may be randomly generated. The position information of each bit value of the binary code may be queried from the vector conversion table to obtain the position vector corresponding to the position information, and then feature fusion may be performed based on each position vector to obtain a fused feature. The specific feature fusion process will be described in detail below.

[0124] The embodiment of the present application determines the binary coding of the similarity between the object feature and the recommended feature through coding conversion, and performs vector mapping and fusion based on the position information of the binary code to obtain the fusion feature of the object feature and the recommended feature; since the above-mentioned similarity can well characterize the correlation between the object feature and the recommended feature, it will not cause the loss of feature information, and can also make the two feature vectors in the same space to better adapt the information recommendation model, effectively improving the efficiency of information recommendation.

[0125] In an embodiment of the present application, a possible implementation method is provided, wherein the feature encoding is performed based on the similarity between the object feature and the recommended feature to obtain the encoding information corresponding to the similarity, including:

[0126] S301, adjusting the value of the similarity based on a preset numerical range to obtain a target similarity.

[0127] In some embodiments, the similarity may be the cosine similarity between the object feature and the recommendation feature. For example, the object feature may be represented by the first vector x u To represent, the recommended feature can be represented by the second vector x a to characterize;

[0128] Among them, x u ={x u1 ,x u2 …x ui …x ud}, x a ={x a1 ,x a2 …x ai …x ad}; The dimensions of the first and second vectors are both d, and the vector at the i-th position in the first vector is x ui , the vector at the i-th position in the second vector is x ai The cosine similarity cos score can be calculated based on the following formula:

[0129]

[0130] The value range of the above similarity cos score is [-1,1].

[0131] Specifically, a preset numerical range can be set based on actual application conditions. For example, the numerical range can be set to [0, 2*10 n ], where n is a positive integer; the similarity can be amplified and rounded based on the following formula (2):

[0132] dist = round(10 n *(1-cos score)) (2)

[0133] Among them, dist is the enlarged target similarity. When the angle between the first vector and the second vector is smaller, the target similarity dist is smaller, and the distance between the recommended feature and the object feature is closer. The above enlargement and rounding of the cosine similarity can lay a good foundation for the subsequent binary coding conversion and control the dimension of the binary coding.

[0134] S302, performing binary conversion on the target similarity to determine the binary code corresponding to the similarity.

[0135] In some implementations, when the target similarity is 50, a binary conversion is performed on it, and the corresponding binary code is 110010; when the target similarity is 51, a binary conversion is performed on it, and the corresponding binary code is 110011.

[0136] The embodiment of the present application amplifies and rounds the similarity between the object feature and the recommended feature through a preset numerical interval to determine the corresponding target similarity; then the target similarity is converted into a binary code, which can achieve the dimensionality unification of the correlation relationship between the object feature and the recommended feature, that is, the similarity, without losing the feature information, laying a good foundation for subsequent feature fusion, and further improving the representation ability of the fused features and the accuracy of information recommendation.

[0137] A possible implementation method is provided in the embodiment of the present application, such as Figure 3 As shown, the position information of each bit value of the binary code is converted into a vector to obtain the position vector corresponding to each bit value, and the position vectors are fused to obtain the fusion feature, including:

[0138] S401, for each bit value, query the position vector corresponding to the value in a preset vector conversion table.

[0139] Among them, the vector table includes the relationship between the position information of the numerical value and the position vector. The position vector corresponding to each bit of the above binary code can be randomly generated, and the generated position vectors are not repeated. For example, the maximum value K of the length of the binary code corresponding to each recommended feature can be determined first, and then K position vectors with a dimension of M can be randomly generated. The first bit (lowest bit) of the binary code corresponds to the position vector y1, the second bit corresponds to the position vector y2, and so on. The Kth bit corresponds to the position vector y k .

[0140] Specifically, when the value in the binary code is 1, the position vector corresponding to the position of the value can be directly obtained; when the value in the binary code is 0, the position vector of the value can be set to 0. Figure 4As shown in the figure, when the target similarity is 50, a binary conversion is performed on it, and the corresponding binary code is 110010. Then, the corresponding position vector can be activated according to the position with a value of 1, that is, the position vector corresponding to the binary code includes y2, y5, and y6; the position vector with a value of 0 is set to 0 and does not participate in subsequent calculations.

[0141] S402, based on the dimension of the position vector, each position vector is added bit by bit to obtain a fusion feature of the object feature and the recommendation feature.

[0142] Specifically, since the dimensions of the position vectors are the same, the position vectors can be added bit by bit to obtain a fusion feature of the object feature and the recommendation feature.

[0143] The embodiment of the present application uses a preset vector conversion table to query the position vector corresponding to each bit value in the binary code, and then adds each position vector bit by bit to obtain the final fusion feature, thereby achieving an effective fusion of the object feature and the recommendation feature. The fusion feature can effectively characterize the correlation between the object feature and the recommendation feature, and can also be adapted to the information recommendation model, effectively improving the pertinence and accuracy of subsequent information matching.

[0144] A possible implementation method is provided in the embodiment of the present application, such as Figure 5 As shown, the above method further includes:

[0145] S501, obtaining object information of a recommended object, numerically encoding the object information, and obtaining candidate object features of the recommended object.

[0146] For example, if the information to be recommended is an advertisement, and the recommended object is a user, the object information may be the user's gender, age, job, etc. The above numerical encoding may include one-hot encoding.

[0147] S502, for each piece of information to be recommended, numerically encode the information to be recommended to obtain candidate recommendation features of the information to be recommended; and concatenate the candidate recommendation features with the candidate object features to obtain concatenated features corresponding to the information to be recommended.

[0148] When the recommendation information is an advertisement, its image size, format, display rules, etc. may be one-hot encoded.

[0149] S503, the above-mentioned inputting each fusion feature into the pre-trained information recommendation model to perform data prediction, and obtaining the matching degree between the recommended object and each information to be recommended, including:

[0150] For each piece of information to be recommended, the corresponding fusion features and concatenation features are input into the pre-trained information recommendation model for data prediction to obtain the matching degree between the recommended object and the information to be recommended.

[0151] Specifically, the detailed steps of the information recommendation model for data prediction will be introduced in detail below.

[0152] A possible implementation method is provided in the embodiment of the present application, such as Figure 6 As shown, the corresponding fusion features and concatenation features are input into the pre-trained information recommendation model for data prediction to obtain the matching degree between the recommended object and the information to be recommended, including:

[0153] S601, stitching the splicing feature and the fusion feature to obtain a fusion splicing feature.

[0154] In the embodiment of the present application, in order to ensure the integrity of feature information, the splicing feature and the fusion feature may be spliced ​​again to obtain a fusion splicing feature.

[0155] S602: Perform data prediction on the fused and spliced ​​features based on the information recommendation model to obtain a matching degree between the recommended object and the information to be recommended.

[0156] Specifically, taking the information recommendation model as an artificial neural network as an example, it may include an input unit, a hidden unit, and an output unit. After receiving the input fusion splicing features, the input unit performs optimization processing such as feature screening and feature cleaning, and inputs the optimized fusion splicing features into the hidden unit. The hidden unit performs feature enhancement processing on the optimized fusion splicing features based on the connection weights between each neuron to obtain enhanced features, and inputs the enhanced features into the output unit; after receiving the enhanced features, the output unit may perform classification processing on the enhanced features based on the classification function, and obtain the matching degree corresponding to the classification result, and use the matching degree as output.

[0157] The embodiment of the present application numerically encodes the recommended object and the information to be recommended respectively to obtain the spliced ​​features of the information to be recommended and the recommended object, and then the spliced ​​features can be secondary spliced ​​with the fused features obtained by fusing the previous features as the input of the information recommendation model to obtain the matching degree between the recommended object and the information to be recommended; since the input of the information recommendation model includes the spliced ​​features and fused features corresponding to the information to be recommended and the recommended object, object features and recommendation features can be obtained from multiple dimensions to achieve accurate data prediction based on the information recommendation model, further improving the effectiveness and accuracy of the matching degree.

[0158] A possible implementation method is provided in the embodiment of the present application, such as Figure 7 As shown in the figure, the above information recommendation model is trained based on the following method:

[0159] S701, obtaining sample object features of a historically recommended object, and obtaining sample recommendation features of a plurality of historically recommended information corresponding to the historically recommended object.

[0160] S702, obtaining access records of historical recommendation information, and generating a tag matching degree between each historical recommendation information and a historical recommendation object based on the access records.

[0161] The access records may include the number of clicks, number of conversions or value, etc., for the historical recommendation information.

[0162] Taking the historical recommendation information as an advertisement as an example, the corresponding access record may specifically include the number of clicks by users on the advertisement, the number of downloads by new users, the number of attention by new users, the submission rate of the corresponding form, and the cost of placing the advertisement.

[0163] Specifically, the server or terminal for information recommendation can collect statistics on the access records to obtain statistical values ​​of different dimensions such as click-through rate, conversion rate and value of the historical recommended information. Then, the statistical values ​​of different dimensions are weighted and summed to obtain the label matching degree of the historical recommended information and the historical recommended object.

[0164] S703, for each sample recommendation feature, determine the sample similarity between the sample object feature and the sample recommendation feature; perform feature encoding based on the sample similarity between the sample object feature and the sample recommendation feature to obtain sample encoding information corresponding to the sample similarity; determine the sample position information of each numerical value in the sample encoding information, and generate a sample fusion feature of the sample object feature and the sample recommendation feature based on the sample position information of each numerical value and the sample encoding information.

[0165] S704: Input each sample fusion feature into a preset initial model for data prediction to obtain a sample matching degree between each historical recommendation information and the historical recommendation object.

[0166] S705, performing at least one update operation on the initial model based on the matching degree of each sample and the matching degree of the corresponding label, until a preset update end condition is met, and using the initial model after the last update as the information recommendation model.

[0167] The update operation includes:

[0168] Calculate the difference between the sample matching degree and the corresponding label matching degree according to the preset loss function;

[0169] Furthermore, the update end condition includes: the difference is not greater than a preset convergence threshold.

[0170] The embodiment of the present application generates sample recommendation features, sample object features, and the label matching degree corresponding to the sample recommendation features through historical recommendation objects and corresponding historical recommendation information. Feature encoding and feature fusion are then performed based on the similarity between the sample recommendation features and the sample object features to obtain sample fusion features, and then the sample fusion features and the initial model preset by the label matching degree input value are trained until the information recommendation model is obtained; the embodiment of the present application improves the representation ability of sample fusion features through similarity calculation, feature encoding, and fusion, effectively improving the efficiency and effect of model training, so that the information recommendation model can fully learn the characteristic information of historical recommendation information and historical recommendation objects, better capture the information preferences of historical recommendation objects, and effectively improve the push pertinence and accuracy of the information recommendation model.

[0171] In order to better understand the above information recommendation method, Figure 8 An example of an information recommendation method of the present application is described in detail. The method can be applied to an information recommendation server, comprising the following steps:

[0172] S801, the information recommendation server obtains the object information of the recommended object and the corresponding at least two pieces of information to be recommended from a preset database; then, based on the word vector model, vector mapping is performed on the above object information and each piece of information to be recommended to obtain the object features of the recommended object and the recommendation features of at least two pieces of information to be recommended.

[0173] Furthermore, the information recommendation server can also numerically encode the object information to obtain candidate object features of the recommended object; and numerically encode each information to be recommended to obtain candidate recommendation features of the information to be recommended; and then splice the candidate recommendation features and the candidate object features to obtain the spliced ​​features corresponding to the information to be recommended.

[0174] S802, for each recommended feature, calculate the cosine value of the angle between the object feature and the recommended feature, and use the cosine value as the similarity between the object feature and the recommended feature; then adjust the value of the similarity based on a preset numerical range to obtain a target similarity.

[0175] S803, perform binary conversion on the target similarity to determine the binary code corresponding to the similarity; then, for each bit value of the binary code, query the position vector corresponding to the value in the preset vector conversion table, and based on the dimension of the position vector, add each position vector bit by bit to obtain the fusion feature of the object feature and each recommended feature.

[0176] S804, stitching the splicing feature and the fusion feature to obtain a fusion splicing feature.

[0177] S805, performing data prediction on the fused and spliced ​​features based on the information recommendation model to obtain a matching degree between the recommended object and the information to be recommended.

[0178] The above-mentioned matching degree may include the click rate, conversion rate or value of the information to be recommended.

[0179] Specifically, taking the information recommendation model as an artificial neural network as an example, it may include an input unit, a hidden unit, and an output unit. After receiving the input fusion splicing features, the input unit performs optimization processing such as feature screening and feature cleaning, and inputs the optimized fusion splicing features into the hidden unit. The hidden unit performs feature enhancement processing on the optimized fusion splicing features based on the connection weights between each neuron to obtain enhanced features, and inputs the enhanced features into the output unit; after receiving the enhanced features, the output unit may perform classification processing on the enhanced features based on the classification function, and obtain the matching degree corresponding to the classification result, and use the matching degree as output.

[0180] S806: The information recommendation server selects target recommendation information from the information to be recommended based on the matching degree and a preset screening threshold, and sends the target recommendation information to the client so that the recommended object obtains the target recommendation information.

[0181] In the embodiment of the present application, taking the case where the information to be recommended is an advertisement and the recommended object is an advertisement position in a web page or a public account as an example, the corresponding matching degree may include the account attention rate, APP download rate, etc. obtained by the advertisement in its advertisement position. The account attention rate and the APP download volume may be weighted and summed based on the preset weights to obtain the final matching degree, and the advertisement corresponding to the final matching degree greater than the preset threshold may be used as the target advertisement, and the target advertisement may be pushed to the advertisement position for display, so as to achieve targeted push based on the advertisement position and improve the conversion rate of the advertisement position.

[0182] The embodiment of the present application obtains the object feature of the recommended object and the recommendation features of at least two pieces of information to be recommended, determines the similarity between the object feature and each recommendation feature, generates a fusion feature of the object feature and each recommendation feature based on the encoding information corresponding to the similarity, and the position information of each numerical value in the encoding information; then inputs each fusion feature into the pre-trained information recommendation model for data prediction, and then obtains the matching degree between the recommended object and each piece of information to be recommended, so as to filter out the target recommendation information from the information to be recommended based on the matching degree, and recommend the target recommendation information to the recommended object. The embodiment of the present application can encode the similarity between the object feature and each recommendation feature, obtain the encoding information corresponding to the similarity, and then perform feature fusion according to the encoding information and the position information of each numerical value in the encoding information, and generate a fusion feature of the object feature and the recommendation feature, the fusion feature can effectively characterize the association between the recommended object and the information to be recommended, and can be well adapted to the information recommendation model, thereby improving the accuracy of the matching degree between the recommended object and the information to be recommended.

[0183] Different from the prior art in which the features of the recommended object and the information to be recommended are discretized to obtain fused features, the feature fusion through similarity encoding in this application can retain the complete original feature information, enable the information recommendation model to fully learn the information preferences of the recommended object, and ensure the accuracy and pertinence of the target recommended information.

[0184] The present application embodiment provides an information recommendation device, such as Fig. 9 As shown, the information recommendation device 90 may include: an acquisition module 901, a fusion module 902, a prediction module 903 and a recommendation module 904;

[0185] The acquisition module 901 is used to acquire the object features of the recommended object and at least two recommendation features of the information to be recommended;

[0186] The fusion module 902 is used to determine the similarity between the object feature and the recommended feature for each recommended feature; perform feature encoding based on the similarity between the object feature and the recommended feature to obtain encoding information corresponding to the similarity; determine the position information of each value in the encoding information, and generate a fusion feature of the object feature and the recommended feature based on the position information of each value and the encoding information;

[0187] Prediction module 903, used to input each fusion feature into the pre-trained information recommendation model for data prediction, and obtain the matching degree between the recommended object and each information to be recommended;

[0188] The recommendation module 904 is used to filter out target recommendation information from the information to be recommended based on the matching degree, and recommend the target recommendation information to the recommended object.

[0189] A possible implementation method is provided in an embodiment of the present application, wherein the above-mentioned encoding information includes binary encoding;

[0190] When generating a fusion feature of the object feature and the recommendation feature based on the position information and the encoding information of each value, the fusion module 902 is used to:

[0191] The position information of each bit value of the binary code is converted into a vector to obtain the position vector corresponding to each bit value, and the position vectors are fused to obtain the fused feature.

[0192] In an embodiment of the present application, a possible implementation method is provided. When the fusion module 902 performs feature encoding based on the similarity between the object feature and the recommended feature to obtain encoding information corresponding to the similarity, it is used to:

[0193] Adjust the value of the similarity based on a preset numerical range to obtain a target similarity;

[0194] Perform binary conversion on the target similarity and determine the binary code corresponding to the similarity.

[0195] A possible implementation method is provided in an embodiment of the present application. The fusion module 902 performs vector conversion on the position information of each bit value of the binary code to obtain the position vector corresponding to each bit value, and fuses the position vectors to obtain the fusion feature, which is used to:

[0196] For each bit of the numerical value, the position vector corresponding to the numerical value is searched in a preset vector conversion table; wherein the vector table includes the correlation between the position information of the numerical value and the position vector;

[0197] Based on the dimension of the position vector, each position vector is added bit by bit to obtain the fusion feature of the object feature and the recommendation feature.

[0198] A possible implementation method is provided in an embodiment of the present application, wherein the acquisition module 901 is further configured to:

[0199] Obtaining object information of the recommended object, numerically encoding the object information, and obtaining candidate object features of the recommended object;

[0200] For each piece of information to be recommended, numerically encode the information to be recommended to obtain candidate recommendation features of the information to be recommended; concatenate the candidate recommendation features with the candidate object features to obtain concatenated features corresponding to the information to be recommended;

[0201] The above prediction module is used to input each fusion feature into the pre-trained information recommendation model for data prediction to obtain the matching degree between the recommended object and each information to be recommended:

[0202] For each piece of information to be recommended, the corresponding fusion features and concatenation features are input into the pre-trained information recommendation model for data prediction to obtain the matching degree between the recommended object and the information to be recommended.

[0203] A possible implementation method is provided in an embodiment of the present application. When the prediction module 903 inputs the corresponding fusion features and concatenation features into the pre-trained information recommendation model for data prediction and obtains the matching degree between the recommended object and the information to be recommended, it is used to:

[0204] The splicing feature and the fusion feature are spliced ​​to obtain the fusion splicing feature;

[0205] Based on the information recommendation model, data prediction is performed on the fused splicing features to obtain the matching degree between the recommended object and the information to be recommended.

[0206] A possible implementation method is provided in an embodiment of the present application. The above information recommendation model is trained based on the following method:

[0207] Obtaining sample object features of a historically recommended object, and obtaining sample recommendation features of a plurality of historically recommended information corresponding to the historically recommended object;

[0208] Obtain access records of historical recommendation information, and generate label matching degrees between each historical recommendation information and the historical recommendation object based on the access records;

[0209] For each sample recommendation feature, determine the sample similarity between the sample object feature and the sample recommendation feature; perform feature encoding based on the sample similarity between the sample object feature and the sample recommendation feature to obtain sample encoding information corresponding to the sample similarity; determine the sample position information of each value in the sample encoding information, and generate a sample fusion feature of the sample object feature and the sample recommendation feature based on the sample position information of each value and the sample encoding information;

[0210] Input each sample fusion feature into the preset initial model for data prediction to obtain the sample matching degree between each historical recommendation information and the historical recommendation object;

[0211] Based on the matching degree of each sample and the matching degree of the corresponding label, the initial model is updated at least once until the preset update end condition is met, and the initial model after the last update is used as the information recommendation model;

[0212] The update operation includes:

[0213] Calculate the difference between the sample matching degree and the corresponding label matching degree according to the preset loss function;

[0214] Update the parameters of the initial model based on the difference;

[0215] The update end condition includes: the difference is not greater than a preset convergence threshold.

[0216] The device of the embodiments of the present application can execute the method provided by the embodiments of the present application, and the implementation principles are similar. The actions performed by each module in the device of each embodiment of the present application correspond to the steps in the method of each embodiment of the present application. For the detailed functional description of each module of the device, please refer to the description in the corresponding method shown in the previous text, which will not be repeated here.

[0217] The embodiment of the present application obtains the object feature of the recommended object and the recommendation features of at least two pieces of information to be recommended, determines the similarity between the object feature and each recommendation feature, generates a fusion feature of the object feature and each recommendation feature based on the encoding information corresponding to the similarity, and the position information of each numerical value in the encoding information; then inputs each fusion feature into the pre-trained information recommendation model for data prediction, and then obtains the matching degree between the recommended object and each piece of information to be recommended, so as to filter out the target recommendation information from the information to be recommended based on the matching degree, and recommend the target recommendation information to the recommended object. The embodiment of the present application can encode the similarity between the object feature and each recommendation feature, obtain the encoding information corresponding to the similarity, and then perform feature fusion according to the encoding information and the position information of each numerical value in the encoding information, and generate a fusion feature of the object feature and the recommendation feature, the fusion feature can effectively characterize the association between the recommended object and the information to be recommended, and can be well adapted to the information recommendation model, thereby improving the accuracy of the matching degree between the recommended object and the information to be recommended. Different from the prior art in which the features of the recommended object and the information to be recommended are discretized to obtain fused features, the feature fusion through similarity encoding in this application can retain the complete original feature information, enable the information recommendation model to fully learn the information preferences of the recommended object, and ensure the accuracy and pertinence of the target recommended information.

[0218] In an embodiment of the present application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory, and the processor executes the above-mentioned computer program to implement the steps of the information recommendation method, which can be implemented compared with the related art: the embodiment of the present application obtains the object feature of the recommended object and the recommendation features of at least two information to be recommended, determines the similarity between the object feature and each recommendation feature, and generates a fusion feature of the object feature and each recommendation feature based on the encoding information corresponding to the similarity and the position information of each value in the encoding information; then each fusion feature is input into the pre-trained information recommendation model for data prediction, and then the matching degree between the recommended object and each information to be recommended is obtained, so as to filter out the target recommendation information from the information to be recommended based on the matching degree, and recommend the target recommendation information to the recommended object. In an embodiment of the present application, the similarity between the object feature and each recommendation feature can be encoded to obtain the encoding information corresponding to the similarity, and then the feature fusion is performed according to the encoding information and the position information of each value in the encoding information, so as to generate a fusion feature of the object feature and the recommendation feature, and the fusion feature can effectively characterize the association relationship between the recommended object and the information to be recommended, and can be well adapted to the information recommendation model, thereby improving the accuracy of the matching degree between the recommended object and the information to be recommended. Different from the prior art in which the features of the recommended object and the information to be recommended are discretized to obtain fused features, the feature fusion through similarity encoding in this application can retain the complete original feature information, enable the information recommendation model to fully learn the information preferences of the recommended object, and ensure the accuracy and pertinence of the target recommended information.

[0219] In an alternative embodiment, an electronic device is provided, such as Fig.10 As shown, Fig.10 The electronic device 100 shown includes: a processor 1001 and a memory 1003. The processor 1001 and the memory 1003 are connected, such as through a bus 1002. Optionally, the electronic device 100 may also include a transceiver 1004, which may be used for data interaction between the electronic device and other electronic devices, such as data transmission and / or data reception. It should be noted that in actual applications, the transceiver 1004 is not limited to one, and the structure of the electronic device 100 does not constitute a limitation on the embodiments of the present application.

[0220] Processor 1001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. Processor 1001 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0221] The bus 1002 may include a path to transmit information between the above components. The bus 1002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus 1002 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig.10 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0222] The memory 1003 may be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compressed optical disk, laser disk, optical disk, digital versatile disk, Blu-ray disk, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium that can be used to carry or store computer programs and can be read by a computer, without limitation herein.

[0223] The memory 1003 is used to store the computer program for executing the embodiment of the present application, and the execution is controlled by the processor 1001. The processor 1001 is used to execute the computer program stored in the memory 1003 to implement the steps shown in the above method embodiment.

[0224] The electronic devices include, but are not limited to, mobile terminals such as mobile phones, notebook computers, PADs, etc. and fixed terminals such as digital TVs, desktop computers, etc.

[0225] An embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps and corresponding contents of the aforementioned method embodiment can be implemented.

[0226] The embodiment of the present application provides a computer program product or a 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 the processor executes the computer instructions, so that when the computer device executes the computer instructions, the following conditions are achieved:

[0227] Obtaining object features of a recommended object and at least two recommendation features of information to be recommended;

[0228] For each recommended feature, determine the similarity between the object feature and the recommended feature; perform feature encoding based on the similarity between the object feature and the recommended feature to obtain encoding information corresponding to the similarity; determine the position information of each value in the encoding information, and generate a fusion feature of the object feature and the recommended feature based on the position information of each value and the encoding information;

[0229] Input each fusion feature into the pre-trained information recommendation model for data prediction to obtain the matching degree between the recommended object and each information to be recommended;

[0230] The target recommended information is filtered out from the information to be recommended based on the matching degree, and the target recommended information is recommended to the recommended object.

[0231] The terms "first", "second", "third", "fourth", "1", "2", etc. (if any) 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 numbers used in this way can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than that shown or described in the drawings.

[0232] It should be understood that, although each operation step is indicated by arrows in the flowchart of the embodiment of the present application, the implementation order of these steps is not limited to the order indicated by the arrows. Unless clearly stated herein, in some implementation scenarios of the embodiment of the present application, the implementation steps in each flowchart can be performed in other orders according to demand. In addition, some or all of the steps in each flowchart may include multiple sub-steps or multiple stages based on actual implementation scenarios. Some or all of these sub-steps or stages may be executed at the same time, and each sub-step or stage in these sub-steps or stages may also be executed at different times respectively. In different scenarios of execution time, the execution order of these sub-steps or stages may be flexibly configured according to demand, and the embodiment of the present application does not limit this.

[0233] The above is only an optional implementation method for some implementation scenarios of the present application. It should be pointed out that for ordinary technicians in this technical field, without departing from the technical concept of the solution of the present application, other similar implementation methods based on the technical ideas of the present application are also within the protection scope of the embodiments of the present application.

Claims

1. An information recommendation method, characterized in that: include: Obtaining object features of a recommended object and at least two recommendation features of information to be recommended; For each recommended feature, determining the similarity between the object feature and the recommended feature; performing feature encoding based on the similarity between the object feature and the recommended feature to obtain encoding information corresponding to the similarity; Determine the position information of each value in the coding information, and generate a fusion feature of the object feature and the recommended feature based on the position information of each value and the coding information; Input each of the fusion features into a pre-trained information recommendation model for data prediction to obtain a matching degree between the recommended object and each piece of information to be recommended; Target recommendation information is screened out from the information to be recommended based on the matching degree, and the target recommendation information is recommended to the recommended object.

2. The method according to claim 1, characterized in that The coded information includes binary code; The generating, based on the position information of each value and the encoding information, a fusion feature of the object feature and the recommendation feature, comprises: The position information of each bit value of the binary code is converted into a vector to obtain a position vector corresponding to each bit value, and the position vectors are fused to obtain the fused feature.

3. The method according to claim 2, characterized in that The performing feature encoding based on the similarity between the object feature and the recommended feature to obtain encoding information corresponding to the similarity includes: Adjusting the value of the similarity based on a preset numerical range to obtain a target similarity; A binary conversion is performed on the target similarity to determine a binary code corresponding to the similarity.

4. The method according to claim 2, characterized in that: The vector conversion is performed on the position information of each bit value of the binary code to obtain the position vector corresponding to each bit value, and the position vectors are fused to obtain the fusion feature, including: For each bit of the numerical value, query the position vector corresponding to the numerical value in a preset vector conversion table; wherein the vector table includes the association relationship between the position information of the numerical value and the position vector; Based on the dimensions of the position vectors, the position vectors are added bit by bit to obtain a fusion feature of the object feature and the recommendation feature.

5. The method according to claim 1, characterized in that The method further comprises: Acquire object information of the recommended object, perform numerical encoding on the object information, and obtain candidate object features of the recommended object; For each piece of information to be recommended, numerically encode the information to be recommended to obtain candidate recommendation features of the information to be recommended; concatenate the candidate recommendation features with the candidate object features to obtain concatenated features corresponding to the information to be recommended; The step of inputting each of the fusion features into a pre-trained information recommendation model for data prediction to obtain a matching degree between the recommended object and each piece of information to be recommended includes: For each piece of information to be recommended, the corresponding fusion feature and concatenation feature are input into a pre-trained information recommendation model for data prediction to obtain a matching degree between the recommended object and the information to be recommended.

6. The method according to claim 5, characterized in that The step of inputting the corresponding fusion features and concatenated features into a pre-trained information recommendation model for data prediction to obtain a matching degree between the recommended object and the information to be recommended includes: Splicing the splicing feature and the fusion feature to obtain a fusion splicing feature; Based on the information recommendation model, data prediction is performed on the fused and spliced ​​features to obtain a matching degree between the recommended object and the information to be recommended.

7. The method according to claim 1, characterized in that The information recommendation model is trained based on the following method: Acquire sample object features of a historically recommended object, and acquire sample recommendation features of a plurality of historical recommendation information corresponding to the historically recommended object; Acquire access records of historical recommendation information, and generate a label matching degree between each historical recommendation information and the historical recommendation object based on the access records; For each sample recommendation feature, determine the sample similarity between the sample object feature and the sample recommendation feature; perform feature encoding based on the sample similarity between the sample object feature and the sample recommendation feature to obtain sample encoding information corresponding to the sample similarity; Determine sample position information of each value in the sample coding information, and generate a sample fusion feature of the sample object feature and the sample recommendation feature based on the sample position information of each value and the sample coding information; Input each of the sample fusion features into a preset initial model for data prediction to obtain a sample matching degree between each historical recommendation information and the historical recommendation object; Based on the matching degree of each sample and the matching degree of the corresponding label, the initial model is updated at least once until a preset update end condition is met, and the initial model after the last update is used as the information recommendation model; The update operation includes: Calculate the difference between the sample matching degree and the corresponding label matching degree according to the preset loss function; updating the parameters of the initial model based on the difference; The update end condition includes: the difference is not greater than a preset convergence threshold.

8. An information recommendation device, characterized in that: include: An acquisition module, used to acquire object features of a recommended object and at least two recommendation features of information to be recommended; A fusion module, configured to determine, for each recommended feature, the similarity between the object feature and the recommended feature; perform feature encoding based on the similarity between the object feature and the recommended feature to obtain encoding information corresponding to the similarity; Determine the position information of each value in the coding information, and generate a fusion feature of the object feature and the recommended feature based on the position information of each value and the coding information; A prediction module, used for inputting each of the fusion features into a pre-trained information recommendation model for data prediction, and obtaining a matching degree between the recommended object and each piece of information to be recommended; The recommendation module is used to filter out target recommendation information from the information to be recommended based on the matching degree, and recommend the target recommendation information to the recommended object.

9. An electronic device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images