A content push system and method based on private domain service

Abstract

The present invention provides a content push system and method based on private domain services, comprising an extraction module, a prediction module, a determination module, a push module, and a private domain service middle platform. The extraction module extracts content feature keywords and user feature keywords from each service content to be recommended. The prediction module applies the content feature keywords and user feature keywords based on a word vector model to obtain a content word vector representation and a user word vector representation. The determination module determines a mixed similarity based on the similarity between the content feature keywords and the user feature keywords, and the similarity between the content word vector representation and the user word vector representation. The push module selects target service content and pushes it to the private domain service user based on the mixed similarity between each service content and the private domain service user. The present invention not only improves the accuracy of push notifications, but also enables precise push notifications based on the personalized needs of private domain service users, providing a more convenient experience for private domain service users.

Description

Technical Field

[0001] The present invention relates to the technical field of private domain services, and in particular to a content push system and method based on private domain services. Background Art

[0002] With the rapid development of the internet, service content resources are becoming increasingly abundant, providing private service users with a wide range of choices. However, faced with this vast amount of service content, private service users often struggle to quickly and accurately find content that meets their needs and interests. Traditional service content push relies primarily on historical behavioral data of private service users or simple tag matching. This approach has limitations in terms of push accuracy and personalization. Therefore, how to efficiently, accurately, and personalized service content push has become a pressing issue. Summary of the Invention

[0003] The present invention provides a content push system and method based on private domain services, which are used to solve the defects in the prior art.

[0004] The present invention provides a content push system based on private domain services, comprising an extraction module, a prediction module, a determination module, a push module, and a private domain service middle platform; the private domain service middle platform is connected to the extraction module, the prediction module, the determination module, and the push module, respectively, to store and manage data of each module;

[0005] An extraction module is used to extract content feature keywords from each service content to be recommended, and to extract user feature keywords from the historical browsing data of private domain service users;

[0006] A prediction module, configured to obtain a content word vector representation and a user word vector representation of each service content based on a word vector model by applying the content feature keywords of each service content and the user feature keywords; the word vector model is obtained by unsupervised training based on text corpus;

[0007] a determination module, configured to determine a mixed similarity between each service content and the private domain service user based on the similarity between the content feature keywords of each service content and the user feature keywords, and the similarity between the content word vector representation of each service content and the user word vector representation;

[0008] The push module is configured to select target service content from each service content based on the mixed similarity between each service content and the private domain service user, and push the target service content to the private domain service user.

[0009] The present invention also provides a content push method based on private domain services, comprising:

[0010] Extracting content feature keywords from each service content to be recommended, and extracting user feature keywords from the historical browsing data of preset private domain service users;

[0011] Based on a word vector model, applying the content feature keywords of each service content and the user feature keywords, a content word vector representation of each service content and a user word vector representation are obtained; the word vector model is obtained by unsupervised training based on text corpus;

[0012] Determining a mixed similarity between each service content and the private domain service user based on the similarity between the content feature keywords of each service content and the user feature keywords, and the similarity between the content word vector representation of each service content and the user word vector representation;

[0013] Based on the mixed similarity between each service content and the private domain service user, target service content is selected from each service content and pushed to the private domain service user.

[0014] According to the content push method based on private domain services provided by the present invention, extracting content feature keywords from each service content includes:

[0015] Determine candidate keywords for each service content;

[0016] Based on the candidate keywords of each service content, a directed weighted graph of each service content is constructed;

[0017] Based on the weight of each node in the directed weighted graph of each service content, a content feature keyword of each service content is selected from the candidate keywords of each service content.

[0018] According to the content push method based on private domain services provided by the present invention, determining candidate keywords for each service content includes:

[0019] Segment the content of each service and perform part-of-speech tagging on each segmented word;

[0020] Based on the part of speech of each word, candidate keywords for each service content are determined.

[0021] According to the content push method based on private domain services provided by the present invention, extracting user characteristic keywords from the historical browsing data of private domain service users includes:

[0022] Determining candidate keywords for the historical browsing data;

[0023] Constructing a directed weighted graph of the private domain service users based on the candidate keywords of the historical browsing data;

[0024] The user characteristic keyword is selected from the candidate keywords in the historical browsing data based on the weight of each node in the directed weighted graph of the private domain service user, the total amount of the historical browsing data, and the amount of the historical browsing data containing the candidate keyword.

[0025] According to the content push method based on private domain services provided by the present invention, the method of selecting the user characteristic keyword from the candidate keywords in the historical browsing data based on the weights of each node in the directed weighted graph of the private domain service user, the total amount of historical browsing data, and the amount of historical browsing data containing the candidate keyword includes:

[0026] Determining a keyword inverse document frequency based on the total amount of historical browsing data and the amount of historical browsing data containing the candidate keyword;

[0027] Determining initial weights of candidate keywords of the historical browsing data based on the keyword inverse document frequency and the weights of each node in the directed weighted graph of the private domain service user;

[0028] Based on the browsing time corresponding to the candidate keyword of the historical browsing data, the earliest browsing time of the historical browsing data, and the latest browsing time of the historical browsing data, the initial weight of the candidate keyword of the historical browsing data is modified to obtain the weight of the candidate keyword of the historical browsing data;

[0029] The user characteristic keyword is selected from the candidate keywords of the historical browsing data based on the weights of the candidate keywords of the historical browsing data.

[0030] According to the content push method based on private domain services provided by the present invention, the word vector model is based on applying the content feature keywords of each service content and the user feature keywords to obtain the content word vector representation and user word vector representation of each service content, including:

[0031] Inputting the content feature keywords of each service content and the user feature keywords into the word vector model respectively to obtain the word vector of the feature keywords of each service content and the word vector of the user feature keywords;

[0032] Based on the weights corresponding to the content feature keywords of each service content, the word vectors of the feature keywords of each service content are weighted and added to obtain the content word vector representation of each service content;

[0033] Based on the weights corresponding to the user feature keywords, the user feature keywords are weighted and added to obtain the user word vector representation.

[0034] According to the content push method based on private domain services provided by the present invention, determining the mixed similarity between each service content and the private domain service user based on the similarity between the content feature keywords of each service content and the user feature keywords, and the similarity between the content word vector representation of each service content and the user word vector representation, includes:

[0035] Based on the preset weights, the similarities between the content feature keywords of each service content and the user feature keywords, as well as the similarities between the content word vector representation of each service content and the user word vector representation are weighted and added to obtain the mixed similarity between each service content and the private domain service user.

[0036] The present invention also provides an electronic device, comprising: a memory for storing a computer software program; a processor for reading and executing the computer software program, thereby implementing any of the above-mentioned content push methods based on private domain services.

[0037] The present invention also provides a non-transitory computer-readable storage medium, in which a computer software program is stored. When the computer software program is executed by a processor, it implements any of the above-mentioned content push methods based on private domain services.

[0038] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned content push methods based on private domain services.

[0039] The content push method based on private domain services provided by the embodiment of the present invention determines the mixed similarity between each service content and the private domain service user based on the similarity between the content feature keywords and user feature keywords of each service content, and the similarity between the content word vector representation and the user word vector representation of each service content, so as to achieve personalized push to the private domain service user based on the mixed similarity. Therefore, it can not only improve the accuracy of push, but also perform precise push according to the personalized needs of the private domain service user, providing a more convenient experience for the private domain service user. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a structural diagram of the content push system based on private domain services provided by the present invention;

[0041] Figure 2 Schematic diagram of the process of the content push method based on private domain service provided by the present invention;

[0042] Figure 3 It is a structural diagram of the CBOW model provided by the present invention;

[0043] Figure 4 Schematic diagram of the framework of the content push method based on private domain service provided by the present invention;

[0044] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0046] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the specified features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0047] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or illustration". Any embodiment of the present invention described as "for example" is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any person skilled in the art to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed herein.

[0048] Reference Figure 1 , Figure 1 This is a schematic diagram of the structure of the private domain service-based content push system provided by the present invention. The private domain service-based content push system includes an extraction module, a prediction module, a determination module, a push module, and a private domain service middleware. The private domain service middleware is connected to the extraction module, prediction module, determination module, and push module respectively to store and manage data from each module.

[0049] Optionally, the extraction module extracts content feature keywords from each service content to be recommended, and extracts user feature keywords from historical browsing data of private domain service users;

[0050] Optionally, the prediction module is based on a word vector model, applying content feature keywords and user feature keywords of each service content to obtain content word vector representation and user word vector representation of each service content; the word vector model is obtained through unsupervised training based on text corpus.

[0051] Optionally, the determination module determines the mixed similarity between each service content and the private domain service user based on the similarity between the content feature keywords and user feature keywords of each service content, and the similarity between the content word vector representation and user word vector representation of each service content.

[0052] Optionally, the push module selects target service content from each service content and pushes it to the private domain service user based on the mixed similarity between each service content and the private domain service user.

[0053] The embodiment of the present invention determines the mixed similarity between each service content and the private domain service user based on the similarity between the content feature keywords and the user feature keywords of each service content, as well as the similarity between the content word vector representation and the user word vector representation of each service content, so as to achieve personalized push to the private domain service user based on the mixed similarity. This not only improves the accuracy of the push, but also enables precise push based on the personalized needs of the private domain service user, providing a more convenient experience for the private domain service user.

[0054] Figure 2 This is a flow chart of the content push method based on private domain service provided by the present invention. Figure 2 As shown, the content push method based on private domain services includes:

[0055] Step 10: extract content feature keywords from each service content to be recommended, and extract user feature keywords from the historical browsing data of private domain service users.

[0056] Here, the service content can be video service content, audio service content, text service content, etc., and the text service content can be virtual service content, such as coupons, shopping vouchers, consumption vouchers, etc. In the case where the service content is video service content, the audio data in the service content is extracted, and the audio data is converted into text data through voice recognition. In the case where the service content is audio service content, the audio data of the audio service content is converted into text data through voice recognition. In the case where the service content is text-visible, text recognition (such as OCR recognition) is used to extract electronic text data.

[0057] Content feature keywords are used to represent the key information in the corresponding service content. Therefore, the TF-IDF value of each word in the service content can be calculated based on TF-IDF (Term Frequency-Inverse Document Frequency). The higher the TF-IDF value, the higher the frequency of occurrence of the corresponding word in the service content, and the greater the probability that the corresponding word is a keyword. Among them, term frequency (TF) indicates the frequency of occurrence in the service content, and inverse document frequency (IDF) indicates the importance of the word in the corpus.

[0058] Furthermore, the historical browsing data of private domain service users can be understood as information about the service content that these users have browsed. User-specific keywords are keywords extracted from this historical browsing data and are used to describe or represent the interests, preferences, or characteristics of private domain service users. These keywords may include content-specific keywords for the service content that they have browsed.

[0059] Optionally, if a private domain service user browses a certain service content frequently, it indicates that the private domain service user has a high degree of interest in the service content. In this case, content feature keywords can be extracted from the service content as user feature keywords.

[0060] Step 20: Based on the word vector model, the content feature keywords and user feature keywords of each service content are applied to obtain the content word vector representation and user word vector representation of each service content; the word vector model is obtained by unsupervised training based on text corpus.

[0061] Specifically, the word vector model is a language model that uses an unsupervised approach to extract low-dimensional word vectors containing semantic information from a large text corpus. Word vectors are not only highly efficient, but also represent the semantic relationships between feature words. Using the text data of service content resources as the corpus, the CBOW (Continuous Bag-of-Word Model) model is used to train word vectors, initially setting the word vector dimension to 300.

[0062] Reference Figure 3 , Figure 3 It is a structural diagram of the CBOW model provided by the present invention, such as Figure 3 As shown in the figure, the CBOW model consists of three layers, namely the input layer, the hidden layer, and the output layer. The input layer inputs the word vector corresponding to the context-related words of a certain feature word and outputs the word vector of the specific word.

[0063] In addition, the content word vector of each service content represents the semantic information used to characterize the corresponding content feature keywords, and the user word vector represents the semantic information used to characterize the user feature keywords.

[0064] Optionally, based on a word vector model, word vectors corresponding to content feature keywords of each service content can be determined, and weighted summation of each word vector can be performed to obtain a content word vector representation of each service content. Based on a word vector model, word vectors corresponding to user feature keywords can be determined, and weighted summation of each word vector can be performed to obtain a user word vector representation.

[0065] Step 30: Based on the similarity between the content feature keywords and user feature keywords of each service content, and the similarity between the content word vector representation and user word vector representation of each service content, determine the mixed similarity between each service content and the private domain service user.

[0066] Specifically, the similarity between the content-specific keywords and user-specific keywords of each service content is used to intuitively represent the keyword matching between each service content and private domain service users from a keyword perspective, which can be understood as a keyword similarity matrix. The similarity between the content word vector representation of each service content and the user word vector representation is used to represent the keyword matching between each service content and private domain service users from a semantic perspective, which can be understood as a word vector similarity matrix.

[0067] Taking into account the independence of keywords, the similarity between content feature keywords and user feature keywords lacks semantic analysis. Therefore, the embodiment of the present invention combines the similarity between the content feature keywords and user feature keywords of each service content, as well as the similarity between the content word vector representation and the user word vector representation of each service content, to determine the mixed similarity between each service content and the private domain service user, so as to more accurately characterize the matching degree between each service content and the private domain service user.

[0068] Step 40: Based on the mixed similarity between each service content and the private domain service user, select target service content from each service content and push it to the private domain service user.

[0069] Specifically, the mixed similarity between each service content and the private domain service user is used to characterize the matching degree between each service content and the private domain service user, and can also be used to characterize the degree of interest of the private domain service user in each service content. The higher the mixed similarity, the higher the degree of interest of the private domain service user in the service content.

[0070] Optionally, the service contents may be sorted in descending order of mixed similarity, and the top k service contents may be pushed to the private domain service users.

[0071] The embodiment of the present invention determines the mixed similarity between each service content and the private domain service user based on the similarity between the content feature keywords and the user feature keywords of each service content, as well as the similarity between the content word vector representation and the user word vector representation of each service content, so as to achieve personalized push to the private domain service user based on the mixed similarity. That is, the embodiment of the present invention can not only improve the accuracy of the push, but also perform precise push according to the personalized needs of the private domain service user, providing a more convenient experience for the private domain service user.

[0072] Based on the above embodiment, content feature keywords are extracted from each service content, including:

[0073] Determine candidate keywords for each service content;

[0074] Based on the candidate keywords of each service content, a directed weighted graph of each service content is constructed;

[0075] Based on the weight of each node in the directed weighted graph of each service content, a content feature keyword of each service content is selected from the candidate keywords of each service content.

[0076] Among them, the candidate keywords for each service content are determined, including:

[0077] Segment the content of each service and perform part-of-speech tagging on each segmented word;

[0078] Based on the part of speech of each word, candidate keywords for each service content are determined.

[0079] As an optional embodiment, firstly, the content text T of each service content is segmented, POS tagged, and stop words are removed, and words with POS such as nouns, verbs, and adjectives are retained. For example, the jieba segmentation can be used, and the POS such as n, nz, v, vd, vn, l, a, and d can be retained as candidate keywords, thus obtaining n candidate keywords, namely, T = *t1, t2, ..., t n +.

[0080] Construct a directed weighted graph G = (V, E) for the candidate keywords of each service content, where V represents the set of nodes, E represents the set of edges, and E is a subset of V × V. n ], and construct an edge between any two nodes using the co-occurrence relationship. An edge exists between two nodes only if their corresponding words co-occur in a window of length K. K represents the window size, that is, at most K words co-occur. i ,V j The weight of the edge between them is w ji , for a given point V i,in(V i ) is the set of points pointing to this point, out(V i ) is point V i The set of points to point to.

[0081] The weight of each node is iteratively calculated according to the following formula until convergence, with the limit value being 0.0001.

[0082]

[0083] Among them, S(V i ) can be set to 1. d is the damping coefficient, ranging from 0 to 1, representing the probability of pointing from a specific point in the graph to any other point. A typical value is 0.85. When using the TextRank algorithm to calculate the scores of each point in the graph, you need to assign an initial value to each point in the graph. This can be any value, and the calculation is performed recursively until convergence.

[0084] According to the weight of each node in the directed weighted graph of each service content, each node is sorted in reverse order, so as to obtain the most important t words as the content feature keywords of each service content.

[0085] On this basis, each service content can be represented as a content feature keyword P of each service content. i :

[0086] P i =( <v i1 ,s i1 >, <v i2 ,s i2 >,…, <v it ,s it >)

[0087] Among them, v ij represents the jth keyword of the i-th service content after reverse sorting, s ij Indicates v ij The corresponding weight.

[0088] Based on any of the above embodiments, extracting user feature keywords from the historical browsing data of private domain service users includes:

[0089] Determine candidate keywords from historical browsing data;

[0090] Based on the candidate keywords of historical browsing data, a directed and authorized graph of private domain service users is constructed;

[0091] Based on the weight of each node in the directed weighted graph of private domain service users, the total amount of historical browsing data, and the amount of historical browsing data containing candidate keywords, user feature keywords are selected from the candidate keywords in the historical browsing data.

[0092] As an optional embodiment, historical browsing data may include service content browsed by private domain service users, and thus the candidate keywords of the private domain service user browsing data may be understood as candidate keywords of the service content browsed by private domain service users, and thus the candidate keywords browsed by private domain service users may be determined with reference to the above-mentioned method for determining candidate keywords of service content, and the embodiments of the present invention will not be repeated here.

[0093] After determining the candidate keywords of the historical browsing data, the directed and authorized graph of the private domain service users can be constructed by referring to the method for constructing the directed and authorized graph of the service content described above.

[0094] Next, based on the weights of each node in the directed weighted graph of the private domain service user, the total amount of historical browsing data, and the amount of historical browsing data containing candidate keywords, user feature keywords are selected from the candidate keywords in the historical browsing data, specifically including:

[0095] Determine the keyword inverse document frequency based on the total number of historical browsing data and the number of historical browsing data containing the candidate keyword;

[0096] Determine the initial weights of candidate keywords for historical browsing data based on the keyword inverse document frequency and the weights of each node in the directed weighted graph of private domain service users;

[0097] Based on the browsing time corresponding to the candidate keyword of the historical browsing data, the earliest browsing time of the historical browsing data, and the latest browsing time of the historical browsing data, the initial weight of the candidate keyword of the historical browsing data is modified to obtain the weight of the candidate keyword of the historical browsing data;

[0098] Based on the weights of the candidate keywords of the historical browsing data, user feature keywords are selected from the candidate keywords of the historical browsing data.

[0099] As an optional embodiment, after determining the candidate keywords of the historical browsing data, the following formula may be used to determine the IDF value corresponding to each candidate keyword:

[0100]

[0101] Among them, |D| represents the total number of historical browsing data of private domain service users, |j:t i ∈d j | indicates that the private domain service user's historical browsing data contains candidate keywords t i The amount of data.

[0102] Next, the initial weight s′ of the candidate keywords of the historical browsing data is determined based on the following formula: ij :

[0103] s′ij =s ij ×IDF j

[0104] Among them, s ij It represents the weight of the jth candidate keyword in the i-th service content browsed by the private domain service user, that is, the weight of the node corresponding to the candidate keyword in the directed weighted graph of the private domain service user.

[0105] As the historical browsing data of private domain service users increases, the new historical browsing data can better reflect the current interest characteristics of private domain service users than the older historical browsing data. In order to more accurately and timely describe user feature keywords, each time the user feature keywords are updated, the exponential forgetting function is used to realize their decay over time. The exponential forgetting function is:

[0106]

[0107] Among them, e -1 <f(t)<1,T max 、T min They represent the latest browsing time and the earliest browsing time of the historical browsing data respectively (which can be calculated on a daily basis).

[0108] Next, based on the following formula, the initial weight of the candidate keywords of the historical browsing data is modified to obtain the weight s″ of the candidate keywords of the historical browsing data ij :

[0109] s″ ij =s′ ij ×f(t)

[0110] Finally, the user characteristic keyword U of the i-th private domain service user i for:

[0111] U i =( <v' i1 ,s″ i1 >, <v' i2 ,s″ i2 >,……, <v' ip ,s ″' ip >)

[0112] Among them, v' ij s″ represents the jth candidate keyword among the candidate keywords extracted from the historical browsing data of the i-th private domain service user, sorted in descending order according to the candidate keyword weights. ij Indicates candidate keyword v' ij The corresponding weight.

[0113] Based on any of the above embodiments, based on the word vector model, the content feature keywords and user feature keywords of each service content are applied to obtain the content word vector representation and user word vector representation of each service content, including:

[0114] Input the content feature keywords and user feature keywords of each service content into the word vector model respectively to obtain the word vector of the feature keyword of each service content and the word vector of the user feature keyword;

[0115] Based on the weights corresponding to the content feature keywords of each service content, the word vectors of the feature keywords of each service content are weighted and added to obtain the content word vector representation of each service content;

[0116] Based on the weights corresponding to the user feature keywords, the user feature keywords are weighted and added to obtain the user word vector representation.

[0117] As an optional embodiment, the content word vector representation of each service content can be determined based on the following formula:

[0118]

[0119] Among them, vec(v ik ) represents the word vector of the kth feature keyword of the i-th service content, s ik Indicates the weight corresponding to the k-th content feature keyword.

[0120] User word vector representation vec(U i ) can be determined based on the following formula:

[0121]

[0122] Among them, vec(v' ik ) represents the word vector of the kth user feature keyword of the i-th private domain service user, s″ ik Indicates the weight corresponding to the k-th user feature keyword.

[0123] Based on any of the above embodiments, determining the mixed similarity between each service content and the private domain service user based on the similarity between the content feature keywords and the user feature keywords of each service content, and the similarity between the content word vector representation and the user word vector representation of each service content, includes:

[0124] Based on the preset weights, the similarities between the content feature keywords and user feature keywords of each service content, as well as the similarities between the content word vector representation and user word vector representation of each service content are weighted and added to obtain the mixed similarity between each service content and the private domain service users.

[0125] In determining the content characteristic keywords P of each service content i And each user feature keyword U i After that, you can use the content feature keywords P of each service content i And each user feature keyword U i Perform similarity calculation to determine the similarity between the content feature keywords of each service content and the user feature keywords. Among them, the cosine matching similarity can be used to determine the similarity between the content feature keywords of each service content and the user feature keywords similar_keyword (P i ,U i ):

[0126]

[0127] Since the similarity between the content feature keywords and the user feature keywords of the above-mentioned service contents is intuitively represented by the keywords of the service content and the private domain service users, the calculation method is simple. However, due to the independence of the keywords, the similarity between the content feature keywords and the user feature keywords of each service content does not take into account the semantic level, and when the number of keywords is relatively large, the sparsity of the vector is more obvious, and the amount of calculation is also significantly increased. Since the word vector has the advantage of representing the semantic information of the keyword, the distribution of semantically similar words in the vector space is also similar. Therefore, the embodiment of the present invention combines the similarity between the content word vector representation of each service content and the user word vector representation to push service content, which can improve the semantic analysis performance and push effect of the push algorithm.

[0128] Among them, the similarity between the content word vector representation of each service content and the user word vector representation is similar_word2vec(vec(P i ), vec(U i )) can be calculated based on the following formula:

[0129] similar_word2vec(vec(P i ), vec(U i ))=cos(vec(P i ), vec(U i ))

[0130] On this basis, based on the preset weights, the similarities between the content feature keywords and user feature keywords of each service content, as well as the similarities between the content word vector representation and the user word vector representation of each service content, are weighted and added to obtain the mixed similarity between each service content and the private domain service user similar(Video(i),User(j)):

[0131] similar(Video(i),User(j))=a×similar_keyword(P i ,U j )+b×similar_word2vec(vec(P i ), vec(U j ))

[0132] Among them, parameter a and parameter b represent weights, with a value range of [0,1]. The initial value can be 0.5, and can be adjusted according to actual conditions in practical applications.

[0133] Based on any of the above embodiments, Figure 4 This is a schematic diagram of the framework of the content push method based on private domain services provided by the present invention. Figure 4 As shown, basic information data of service content is obtained, and the audio corresponding to the service content is converted into speech to obtain the corresponding content text. Feature keywords are extracted from the basic information data of service content and the service content to obtain content feature keywords.

[0134] Obtain historical browsing data of private domain service users (including basic information data of private domain service users and historical records of private domain service users), extract feature keywords from the historical browsing data, and obtain user feature keywords.

[0135] Based on the word vector model, content feature keywords and user feature keywords are applied to obtain content word vector representation and user word vector representation. The word vector model is obtained by preprocessing the text in the corpus and then performing unsupervised training based on the preprocessed text corpus.

[0136] Next, based on the similarity between content feature keywords and user feature keywords (i.e., keyword similarity), and the similarity between content word vector representation and user word vector representation (i.e., word vector similarity), the mixed similarity between service content and private domain service users is determined.

[0137] The service contents are sorted in descending order of the mixed similarity between the service contents and the private domain service users, and a service content recommendation list is generated according to the sorting, so as to push the service contents to the private domain service users according to the recommendation list.

[0138] Figure 5 Schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 5As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other via the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the content push method based on the private domain service, which includes:

[0139] Extracting content feature keywords from each service content to be recommended, and extracting user feature keywords from the historical browsing data of preset private domain service users;

[0140] Based on a word vector model, applying the content feature keywords of each service content and the user feature keywords, a content word vector representation of each service content and a user word vector representation are obtained; the word vector model is obtained by unsupervised training based on text corpus;

[0141] Determining a mixed similarity between each service content and the private domain service user based on the similarity between the content feature keywords of each service content and the user feature keywords, and the similarity between the content word vector representation of each service content and the user word vector representation;

[0142] Based on the mixed similarity between each service content and the private domain service user, target service content is selected from each service content and pushed to the private domain service user.

[0143] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software function module and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0144] On the other hand, the present invention further provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the content push method based on the private domain service provided by the above methods, which includes:

[0145] Extracting content feature keywords from each service content to be recommended, and extracting user feature keywords from the historical browsing data of preset private domain service users;

[0146] Based on a word vector model, applying the content feature keywords of each service content and the user feature keywords, a content word vector representation of each service content and a user word vector representation are obtained; the word vector model is obtained by unsupervised training based on text corpus;

[0147] Determining a mixed similarity between each service content and the private domain service user based on the similarity between the content feature keywords of each service content and the user feature keywords, and the similarity between the content word vector representation of each service content and the user word vector representation;

[0148] Based on the mixed similarity between each service content and the private domain service user, target service content is selected from each service content and pushed to the private domain service user.

[0149] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for pushing content based on private domain services provided by the above methods is implemented, and the method includes:

[0150] Extracting content feature keywords from each service content to be recommended, and extracting user feature keywords from the historical browsing data of preset private domain service users;

[0151] Based on a word vector model, applying the content feature keywords of each service content and the user feature keywords, a content word vector representation of each service content and a user word vector representation are obtained; the word vector model is obtained by unsupervised training based on text corpus;

[0152] Determining a mixed similarity between each service content and the private domain service user based on the similarity between the content feature keywords of each service content and the user feature keywords, and the similarity between the content word vector representation of each service content and the user word vector representation;

[0153] Based on the mixed similarity between each service content and the private domain service user, target service content is selected from each service content and pushed to the private domain service user.

[0154] The system embodiments described above are merely illustrative. Modules described as separate components may or may not be physically separate, and components shown as modules may or may not be physical modules. They may be located in one place or distributed across multiple network modules. Some or all of these modules may be selected based on actual needs to achieve the objectives of this embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0155] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus the necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or certain parts of the embodiment.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A content push system based on private domain services, characterized in that: The content push system based on private domain services includes an extraction module, a prediction module, a determination module, a push module and a private domain service middle platform; the private domain service middle platform is connected to the extraction module, the prediction module, the determination module and the push module respectively to store and manage the data of each module; An extraction module is used to extract content feature keywords from each service content to be recommended, and to extract user feature keywords from the historical browsing data of private domain service users; A prediction module, configured to obtain a content word vector representation and a user word vector representation of each service content based on a word vector model by applying the content feature keywords of each service content and the user feature keywords; the word vector model is obtained by unsupervised training based on text corpus; a determination module, configured to determine a mixed similarity between each service content and the private domain service user based on the similarity between the content feature keywords of each service content and the user feature keywords, and the similarity between the content word vector representation of each service content and the user word vector representation; a push module, configured to select target service content from each service content and push it to the private domain service user based on the mixed similarity between each service content and the private domain service user; The extraction of content feature keywords from each service content includes: Determine candidate keywords for each service content; Based on the candidate keywords of each service content, a directed weighted graph of each service content is constructed; Selecting content feature keywords for each service content from candidate keywords for each service content based on the weights of each node in the directed weighted graph of each service content; The candidate keywords for determining each service content include: Segment the content of each service and perform part-of-speech tagging on each segmented word; Determine candidate keywords for each service content based on the part of speech of each segmented word; The weight of each node is iteratively calculated according to the following formula until convergence, with the limit value being 0.0001; Among them, S(V i ) The initial value can be set to 1; d is the damping coefficient, ranging from 0 to 1, representing the probability of pointing from a specific point in the graph to any other point; any two points V in the directed weighted graph i ,V j The weight of the edge between them is w ji ; For a given point V i ,in(V i ) is the point V i The point set, out(V i ) is point V i The set of points pointed to; Extracting user feature keywords from the historical browsing data of private domain service users includes: Determining candidate keywords for the historical browsing data; Constructing a directed weighted graph of the private domain service users based on the candidate keywords of the historical browsing data; Selecting the user characteristic keyword from the candidate keywords in the historical browsing data based on the weight of each node in the directed weighted graph of the private domain service user, the total amount of the historical browsing data, and the amount of the historical browsing data containing the candidate keyword; The selecting the user characteristic keyword from the candidate keywords in the historical browsing data based on the weight of each node in the directed weighted graph of the private domain service user, the total amount of the historical browsing data, and the amount of the historical browsing data containing the candidate keyword includes: Determining a keyword inverse document frequency based on the total amount of historical browsing data and the amount of historical browsing data containing the candidate keyword; Determining initial weights of candidate keywords of the historical browsing data based on the keyword inverse document frequency and the weights of each node in the directed weighted graph of the private domain service user; Based on the browsing time corresponding to the candidate keyword of the historical browsing data, the earliest browsing time of the historical browsing data, and the latest browsing time of the historical browsing data, the initial weight of the candidate keyword of the historical browsing data is modified to obtain the weight of the candidate keyword of the historical browsing data; Selecting the user characteristic keyword from the candidate keywords of the historical browsing data based on the weights of the candidate keywords of the historical browsing data; The following formula is used to determine the IDF value corresponding to each candidate keyword: i : Among them, |D| represents the total number of historical browsing data of private domain service users, |j:t i ∈d j | indicates that the private domain service user's historical browsing data contains candidate keywords t i The amount of data; The word vector model is based on applying the content feature keywords of each service content and the user feature keywords to obtain the content word vector representation and the user word vector representation of each service content, including: Inputting the content feature keywords of each service content and the user feature keywords into the word vector model respectively to obtain the word vector of the feature keywords of each service content and the word vector of the user feature keywords; Based on the weights corresponding to the content feature keywords of each service content, the word vectors of the feature keywords of each service content are weighted and added to obtain the content word vector representation of each service content; Based on the weights corresponding to the user feature keywords, the weights of the user feature keywords are added to obtain the user word vector representation; Determining the mixed similarity between each service content and the private domain service user based on the similarity between the content feature keywords of each service content and the user feature keywords, and the similarity between the content word vector representation of each service content and the user word vector representation, includes: Based on the preset weights, the similarities between the content feature keywords of each service content and the user feature keywords, as well as the similarities between the content word vector representation of each service content and the user word vector representation are weighted and added to obtain the mixed similarity between each service content and the private domain service user.

2. A content push method based on private domain services, characterized in that: include: Extracting content feature keywords from each service content to be recommended, and extracting user feature keywords from the historical browsing data of preset private domain service users; Based on a word vector model, applying the content feature keywords of each service content and the user feature keywords, a content word vector representation of each service content and a user word vector representation are obtained; the word vector model is obtained by unsupervised training based on text corpus; Determining a mixed similarity between each service content and the private domain service user based on the similarity between the content feature keywords of each service content and the user feature keywords, and the similarity between the content word vector representation of each service content and the user word vector representation; Based on the mixed similarity between each service content and the private domain service user, selecting a target service content from each service content and pushing it to the private domain service user; The extraction of content feature keywords from each service content includes: Determine candidate keywords for each service content; Based on the candidate keywords of each service content, a directed weighted graph of each service content is constructed; Selecting content feature keywords for each service content from candidate keywords for each service content based on the weights of each node in the directed weighted graph of each service content; The candidate keywords for determining each service content include: Segment the content of each service and perform part-of-speech tagging on each segmented word; Determine candidate keywords for each service content based on the part of speech of each segmented word; The weight of each node is iteratively calculated according to the following formula until convergence, with the limit value being 0.0001; Among them, S(V i ) The initial value can be set to 1; d is the damping coefficient, ranging from 0 to 1, representing the probability of pointing from a specific point in the graph to any other point; any two points V in the directed weighted graph i ,V j The weight of the edge between them is w ji ; For a given point V i ,in(V i ) is the point V i The point set, out(V i ) is point V i The set of points pointed to; Extracting user feature keywords from the historical browsing data of private domain service users includes: Determining candidate keywords for the historical browsing data; Constructing a directed weighted graph of the private domain service users based on the candidate keywords of the historical browsing data; Selecting the user characteristic keyword from the candidate keywords in the historical browsing data based on the weight of each node in the directed weighted graph of the private domain service user, the total amount of the historical browsing data, and the amount of the historical browsing data containing the candidate keyword; The selecting the user characteristic keyword from the candidate keywords in the historical browsing data based on the weight of each node in the directed weighted graph of the private domain service user, the total amount of the historical browsing data, and the amount of the historical browsing data containing the candidate keyword includes: Determining a keyword inverse document frequency based on the total amount of historical browsing data and the amount of historical browsing data containing the candidate keyword; Determining initial weights of candidate keywords of the historical browsing data based on the keyword inverse document frequency and the weights of each node in the directed weighted graph of the private domain service user; Based on the browsing time corresponding to the candidate keyword of the historical browsing data, the earliest browsing time of the historical browsing data, and the latest browsing time of the historical browsing data, the initial weight of the candidate keyword of the historical browsing data is modified to obtain the weight of the candidate keyword of the historical browsing data; Selecting the user characteristic keyword from the candidate keywords of the historical browsing data based on the weights of the candidate keywords of the historical browsing data; The following formula is used to determine the IDF value corresponding to each candidate keyword: i : Among them, |D| represents the total number of historical browsing data of private domain service users, |j:t i ∈d j | indicates that the private domain service user's historical browsing data contains candidate keywords t i The amount of data; The word vector model is based on applying the content feature keywords of each service content and the user feature keywords to obtain the content word vector representation and the user word vector representation of each service content, including: Inputting the content feature keywords of each service content and the user feature keywords into the word vector model respectively to obtain the word vector of the feature keywords of each service content and the word vector of the user feature keywords; Based on the weights corresponding to the content feature keywords of each service content, the word vectors of the feature keywords of each service content are weighted and added to obtain the content word vector representation of each service content; Based on the weights corresponding to the user feature keywords, the weights of the user feature keywords are added to obtain the user word vector representation; Determining the mixed similarity between each service content and the private domain service user based on the similarity between the content feature keywords of each service content and the user feature keywords, and the similarity between the content word vector representation of each service content and the user word vector representation, includes: Based on the preset weights, the similarities between the content feature keywords of each service content and the user feature keywords, as well as the similarities between the content word vector representation of each service content and the user word vector representation are weighted and added to obtain the mixed similarity between each service content and the private domain service user.

3. An electronic device, characterized in that: include: Memory for storing computer software programs; The processor is used to read and execute the computer software program, thereby implementing the content push method based on private domain service as claimed in claim 2.

4. A non-transitory computer-readable storage medium, characterized in that The storage medium stores a computer software program, which, when executed by the processor, implements the content push method based on private domain services as claimed in claim 2.

Images