Information pushing method and device, electronic equipment and storage medium

After the target event occurs on the application platform, the target operation object information is determined based on user behavior data and the low click-through rate and conversion rate caused by the historical behavior of information push is solved, and the relevance of information is improved and accuracy.

CN119066255AActive Publication Date: 2024-12-03BEIJING WODONG TIANJUN INFORMATION TECH CO LTD +1
View PDF 12 Cites 0 Cited by

Patent Information

Application Number
CN202310638740.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-31
Publication Date
2024-12-03
Estimated Expiration
2043-05-31

AI Technical Summary

Technical Problem

In the prior art, information push relies on user historical behavior characteristics, resulting in low click-through rate and conversion rate of information push, and users are prone to fatigue.

Method used

After the target event occurs on the application platform, user behavior data is obtained, target search terms associated with the target event are filtered out, keywords are extracted, trigger words are determined, and target operation object information is obtained based on the user's target operation behavior for pushing. Vectorized characterization and vector search libraries improve the relevance and accuracy of information.

Benefits of technology

It improves the click-through rate and conversion rate of information push, avoids the fatigue of information push, ensures that the push information is highly correlated with the target event, and enhances the accuracy and timeliness of information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066255A_ABST
    Figure CN119066255A_ABST
Patent Text Reader

Abstract

The invention provides an information pushing method and device, electronic equipment and a storage medium. Relates to the technical field of information processing. The method comprises the steps of obtaining user behavior data of an application platform after determining that a target event occurs on the application platform; screening out a target search word associated with the target event from the user behavior data; performing keyword extraction on the target search word, and determining a trigger word corresponding to the target event; obtaining a target operation behavior after the user searches the trigger word within the target deadline, and determining target operation object information according to the target operation behavior; and pushing the target operation object information to the application platform as push information. The information pushing method and device can solve the problem that the click rate and the conversion rate of the pushed information are low due to the fact that the information is pushed only depending on the historical behavior characteristics of the user in the related technology.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

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

[0002] With the continuous development and application of Internet technology, excessive information volume has become a major challenge for users to process information. Therefore, major platforms will recommend personalized information to users to alleviate the trouble faced by users with large amounts of information, which can also increase the accuracy of information push.

[0003] In the related technologies, personalized information is mostly pushed to users based on the historical behavior characteristics of users. However, it is difficult to introduce new information into the information pushed in this way, which can easily make users feel tired of the information, thereby affecting the click rate and conversion rate of the pushed information.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0005] The purpose of the embodiments of the present disclosure is to provide an information push method and device, an electronic device, and a storage medium, thereby solving to a certain extent the problem of low click-through rate and conversion rate of pushed information caused by the related technology that completely relies on the user's historical behavior characteristics to push information.

[0006] According to the first aspect of the present disclosure, there is provided an information push method, comprising: after determining that a target event has occurred on an application platform, obtaining user behavior data of the application platform; screening out target search terms associated with the target event from the user behavior data; performing keyword extraction on the target search terms to determine a trigger term corresponding to the target event; obtaining a target operation behavior after the user searches for the trigger term within a target period, and determining target operation object information based on the target operation behavior; and pushing the target operation object information to the application platform as push information.

[0007] Optionally, the method further includes: monitoring user behavior data of the application platform, the user behavior data including a search behavior growth rate; and determining that the target event has occurred in response to the search behavior growth rate for a candidate search term being greater than a preset first threshold.

[0008] Optionally, before pushing the target operation object information, the method also includes: vectorizing the target operation object information; performing a similar vector search in a pre-established vector retrieval library based on the vectorization representation result of the target operation object information to obtain a similar object vector representation; and using the similar object information corresponding to the similar object vector representation as the push information.

[0009] Optionally, before performing keyword extraction on the target search term, the method further includes: performing vectorization processing on the target search term to obtain a target search term vector; performing similar semantic feature extraction on the target search term vector to obtain similar search terms.

[0010] Optionally, the keyword extraction of the target search word includes: taking the similar search words and the target search words as sentences to be extracted; performing keyword extraction on the sentences to be extracted to obtain target keywords; taking the similar search words and the target keywords as trigger words corresponding to the target event.

[0011] Optionally, the user behavior data includes user search terms, and filtering out target search terms associated with the target event from the user behavior data includes: configuring an associated word mapping table for the target event; and filtering out the target search terms from the user search terms according to the associated word mapping table.

[0012] Optionally, the target operation behavior includes a click operation behavior and an order operation behavior, and determining the target operation object information based on the target operation behavior includes: filtering out the click object information corresponding to the click operation behavior within a target time range from the user behavior data, the target time range being a preset time range adjacent to a search time point, and the search time point being a time point when the user searches for the trigger word; filtering out the order object information corresponding to the order operation behavior from the user behavior data; and using the click object information and the order object information as target operation object information.

[0013] Optionally, the target operation object information includes target objects and corresponding target categories arranged in chronological order of operation behaviors, and the vectorized representation of the target operation object information includes: using a trained word vector generation model to respectively vectorize the target objects and corresponding target categories.

[0014] Optionally, the method also includes: obtaining an operation object sequence data set within a specified time range as training sample data; an operation object sequence as a training sample; an operation object in a training sample as an input word, and adjacent operation objects of the operation object as output word labels; inputting each of the training samples into the word vector generation model to obtain a word vector of the corresponding output word; updating the parameters of the word vector generation model with the goal of maximizing the log-likelihood function between the input word and the corresponding output word to train the word vector generation model.

[0015] Optionally, the method also includes: obtaining a candidate object data set based on click data and order data within a specified time range; using a trained word vector generation model to vectorize the candidate object data set to obtain a candidate object vector representation result; and establishing a vector retrieval library based on the candidate object vector representation result.

[0016] Optionally, the method further includes: clustering the trigger words to obtain multiple scene categories.

[0017] The acquiring of training sample data includes: acquiring, according to the multiple scene categories, a sequence data set of operation objects under each scene category as training sample data.

[0018] Optionally, the target operation object information includes the target item and its target category. Before vectorizing the target operation object information, the method also includes: determining a dirty data category; the dirty data category is used to indicate an object category whose object click frequency is greater than a preset second threshold or an object category whose residence time on the object details page is less than a preset third threshold; filtering out object information in the target operation object information that corresponds to the dirty data category and preset sensitive categories and long-tail categories; filtering out delisted object information in the target operation object information.

[0019] According to a second aspect of the present disclosure, there is provided an information push device, which includes: an acquisition module, an extraction module, a determination module and a push module, the acquisition module is used to acquire user behavior data of the application platform after determining that a target event has occurred on the application platform; the extraction module is used to screen out target search terms associated with the target event from the user behavior data; perform keyword extraction on the target search terms to obtain trigger words corresponding to the target event; the determination module is used to acquire target operation behaviors after users search for the trigger words within a target period, and determine target operation object information based on the target operation behaviors; the push module is used to push the target operation object information to the application platform as push information.

[0020] According to a third aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method described in any one of the above embodiments is implemented.

[0021] According to a fourth aspect of the present disclosure, an electronic device is provided, comprising: one or more processors; and a storage device for one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the method described in any one of the above embodiments.

[0022] The exemplary embodiments of the present disclosure may have some or all of the following beneficial effects:

[0023] In the information push method provided by the exemplary embodiment of the present disclosure, on the one hand, the target search words associated with the target event can be screened out from the user behavior data based on the target event, and the push information is determined on this basis. New data different from the user's historical behavior data is introduced through the target event, which avoids the phenomenon that the push information is completely dependent on the user's historical behavior data, thereby solving the problem of fatigue brought to users by the push information generated based on the historical behavior data and the low click-through rate and conversion rate of the push information generated thereby. On the other hand, the target search words associated with the target event can be extracted by keywords to determine the trigger words corresponding to the target event; then the target operation behavior and the corresponding target operation object information after the user searches for the trigger words within the target period are obtained and pushed. According to the target operation behavior after the trigger word search action, the push information is screened and determined, which can ensure that the push information is highly related to the target event, while ensuring the accuracy of the push information, and further ensuring the conversion rate of the push information. In addition, the acquisition of user behavior data and target operation behavior can be acquired instantly to improve the timeliness of the acquired data and ensure the accuracy of the push information.

[0024] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0026] Figure 1 An exemplary application scenario architecture diagram of an information push method and device according to an embodiment of the present disclosure is schematically shown.

[0027] Figure 2 The flowchart of the information push method according to one embodiment of the present disclosure is schematically shown.

[0028] Figure 3 The flowchart schematically shows a process of determining a trigger word corresponding to a target event according to an embodiment of the present disclosure.

[0029] Figure 4 The flowchart schematically shows a process of determining target operation object information according to an embodiment of the present disclosure.

[0030] Figure 5 The flowchart schematically shows a process of determining similar object information according to an embodiment of the present disclosure.

[0031] Figure 6 The flowchart of the vector search library construction process according to one embodiment of the present disclosure is schematically shown.

[0032] Figure 7 The figure schematically shows a processing flow chart of an information push method according to an embodiment of the present disclosure.

[0033] Figure 8 The structural block diagram of an information push device according to an embodiment of the present disclosure is schematically shown.

[0034] Fig. 9 A block diagram of an electronic device suitable for implementing the embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0035] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as being limited to the examples set forth herein; on the contrary, these embodiments are provided so that the present disclosure will be more comprehensive and complete, and the concepts of the example embodiments are fully conveyed to those skilled in the art. The described features, structures, or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known technical solutions are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0036] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0037] like Figure 1 As shown, an exemplary system diagram of an application scenario of an information push method and device is provided, and the system includes a terminal and a server. The method of this embodiment can be applied to a server. It can be understood that the method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. Among them, the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms, or a node in a blockchain. For example, in the process of recommending and selling goods by an e-commerce company, the server can be a server corresponding to an offline system, and the terminal can obtain the push information of the offline system by calling the data repository of the offline system. The terminal can be a smart phone, a tablet computer, a laptop, a desktop computer, a smart speaker, a smart watch, a car-mounted device, etc., but is not limited thereto. When the information push method provided in this embodiment is implemented through the interaction between a terminal and a server, the terminal and the server can be directly or indirectly connected by wired or wireless communication, and the present disclosure is not limited here.

[0038] The information push method provided in the embodiment of the present disclosure may be executed on a server, and accordingly, the information push device is generally disposed in a working server.

[0039] The information push method disclosed in the embodiments of this specification is introduced below in conjunction with specific embodiments.

[0040] refer to Figure 2 As shown, an information push method according to an example implementation provided by the present disclosure may include the following steps.

[0041] Step S210, after determining that a target event occurs on the application platform, obtain user behavior data of the application platform.

[0042] Step S220, screen out target search terms associated with the target event from the user behavior data; perform keyword extraction on the target search terms to determine trigger words corresponding to the target event.

[0043] Step S230, obtaining the target operation behavior after the user searches for the trigger word within the target period, and determining the target operation object information according to the target operation behavior.

[0044] Step S240: Push the target operation object information to the application platform.

[0045] In the information push method provided in this example implementation, on the one hand, the target search terms associated with the target event can be screened out from the user behavior data based on the target event, and the push information can be determined on this basis. New data different from the user's historical behavior data is introduced through the target event, avoiding the phenomenon that the push information is completely dependent on the user's historical behavior data, thereby solving the problem of fatigue brought to users by the push information generated based on historical behavior data and the resulting low click-through rate and conversion rate of the push information. On the other hand, keyword extraction can be performed on the target search terms associated with the target event to determine the trigger words corresponding to the target event; then the target operation behavior and the corresponding target operation object information after the user searches for the trigger words within the target period are obtained and pushed. It can ensure that the push information is highly relevant to the target event, while ensuring the accuracy of the push information, and further ensure the conversion rate of the push information. In addition, the acquisition of user behavior data and target operation behavior can be instant acquisition to improve the timeliness of data acquisition and ensure the accuracy of push information.

[0046] The following describes each step of the present disclosure in more detail.

[0047] In step S210, after determining that a target event occurs on the application platform, user behavior data of the application platform is obtained.

[0048] In this example implementation, the target event refers to a current hot event, such as a hot summer event, a Mid-Autumn Festival event, a Spring Festival event, etc. A hot event may include multiple hot words related to the hot spot. For example, hot words related to the Mid-Autumn Festival event may include "Mid-Autumn Festival", "moon cake", "crab", "lantern", etc. It may be set that when the search behavior of a hot word related to a certain hot spot soars (i.e., grows rapidly in a short period of time), it is determined that a hot event is captured, i.e., the target event occurs.

[0049] In this example implementation, the application platform can be loaded on the user terminal, for example, an electronic mall of an e-commerce platform, a video service application platform, a banking service application platform, etc., which is not limited in this example.

[0050] In this example implementation, user behavior data refers to user behavior characteristics in a specific application scenario. For different application scenarios, user behavior characteristics are different. For example, for the application scenario where users use an e-commerce mall, user behavior characteristics include users' search, browsing, clicking, adding to cart, collecting, ordering, and evaluation of information in the e-commerce mall; for the application scenario where users use video services, user behavior characteristics include users' search, browsing, evaluation, liking, and collection of video information. User behavior data refers to data related to user behavior characteristics and their operation objects, such as search terms, evaluation content, the number, frequency, and growth rate of user behavior characteristics, and the name, sku, and code of the operation object (such as a product or video).

[0051] Exemplarily, user behavior data of the application platform (such as a search behavior growth rate) may be monitored, and in response to a search behavior growth rate for a candidate search term being greater than a preset first threshold, it is determined that a target event has occurred.

[0052] In this example implementation, the candidate search term refers to a hot word corresponding to the target hot event. The search behavior surge is determined by the growth rate of the search behavior for the candidate search term being greater than the preset first threshold, that is, the target event occurs. The relationship between the hot event and the corresponding candidate search term can be configured in advance. The first threshold can be set according to the actual situation, for example, it can be set to a few hours, a day or several days, etc., which is not limited in this example.

[0053] In step S220, target search terms associated with the target event are screened out from the user behavior data; keyword extraction is performed on the target search terms to determine the trigger words corresponding to the target event.

[0054] In this example implementation, the search term refers to the search information input by the user on the application platform. The target search term refers to the search term associated with the target event.

[0055] Exemplarily, a related word mapping table of the target event may be preconfigured; and the target search words are screened out from the user search words according to the related word mapping table.

[0056] In this example, the associated word mapping table is used to give the mapping relationship between the target event and its associated words. By setting the matching rules between the user search words and the associated words, the matching results of the two can be determined according to the matching rules, and the target search words can be screened out according to the matching results. The user search words containing the associated words can also be determined as the target search words, which is not limited in this example. The associated word mapping table can be expanded or deleted according to the actual situation, which is not limited in this example.

[0057] For example, when the target event is a Mid-Autumn Festival event, the corresponding associated words in the associated word mapping table may include "Mid-Autumn Festival", "moon cakes", "lanterns", "crabs", etc. A target event can correspond to one or more associated words. Then match the associated words from the user search words. When the match is successful, the corresponding search word is the target search word. Matching rules can be set, such as calculating the similarity between the associated words and the user search words, and selecting the ones with similarity greater than the similarity threshold to consider the match successful. It can also be set that a specified number of user search words with similarity greater than the similarity threshold are matched successfully, and other matching rules can also be set, which are not limited in this example.

[0058] In some embodiments, keyword extraction may be performed on the target search term, and the extraction result may be used as a trigger word corresponding to the target event.

[0059] In this example implementation, an existing keyword extraction algorithm may be used for keyword extraction, such as the TF-IDF (Term-Frequency Inverse Document Frequency) algorithm, the TextRank algorithm, or a combination of the two. The TextRank algorithm is used as an example for introduction.

[0060] Treat the target search words as sentences, perform word segmentation and part-of-speech tagging on each sentence, and filter out stop words, retaining only words with specified parts of speech, such as nouns, verbs, and adjectives. Use the filtered words as candidate keywords and construct a candidate keyword graph G = (V, E), where V is a node set, such as Figure 3 In TextRank, A, B, C, D, and F are nodes, and E is the edge between nodes (i.e. Figure 3 The edges between nodes in the network are constructed by using the co-occurrence relationship, where each candidate keyword corresponds to a node. The edges between any two nodes are constructed by using the co-occurrence relationship. The edges between two nodes are determined to exist only when the candidate keywords corresponding to the two nodes co-occur in a window of length K. K represents the window size, that is, at most K words co-occur, and K is an integer. The weights of each node are propagated iteratively until convergence. The weights of each node are sorted in reverse order, and the first M candidate keywords are selected as the extraction results.

[0061] In some embodiments, reference Figure 3 , you can determine the trigger word corresponding to the target event through the following steps.

[0062] The first step is to vectorize the target search term to obtain the target search term vector.

[0063] In this example, one-hot vectorization can be used to vectorize the target search term.

[0064] In the second step, similar semantic features are extracted from the target search word vector to obtain similar search words.

[0065] In this example, a pre-trained Bert model may be used to extract similar semantic features from the target search word vector, that is, the target search word vector is input into the Bert model, and similar search words of the input target search word are output.

[0066] The third step is to use similar search terms and target search terms as sentences to be extracted.

[0067] The fourth step is to extract keywords from the sentences to be extracted and obtain the target keywords.

[0068] The fifth step is to use similar search terms and target keywords as trigger words corresponding to the target event.

[0069] In the above embodiments, Figure 3 As shown in the figure, the target search term and the output of the Bert model (similar search terms) can be used as sentences to be extracted, and the TextRank algorithm can be applied to extract keywords from the sentences to be extracted, and the extraction results (target keywords) can be used as trigger terms corresponding to the target event. In this example, the Bert model can be used to expand the target search term, dig out search terms similar to it, and then combine it with the TextRank algorithm to ensure the breadth and accuracy of the determined trigger terms.

[0070] In step S230, the target operation behavior after the user searches for the trigger word within the target period is obtained, and the target operation object information is determined according to the target operation behavior.

[0071] In this example implementation, the target deadline refers to the time period related to the target event, that is, the time period corresponding to the hot event. The target deadline can be set according to the target event. For example, for the Mid-Autumn Festival hot event, the 7-10 days before the Mid-Autumn Festival can be used as the current target deadline. The target operation behavior is used to indicate the operation behavior of the user's interest in the operation object. Different target operator behaviors correspond to different application scenarios. For example: for the e-commerce mall application scenario, the target operation behavior may include operations such as clicking, adding to a purchase, and placing an order. The operation object information refers to the attribute feature information of the operation object. The attribute feature information may include the identification information, name, code, content, etc. of the operation object, such as the product sku (sku is the unique identifier of the product). For the consultation recommendation application scenario, the target operation behavior may include operations such as clicking, collecting, and placing an order. The operation object information may include information such as the code and identification of the consultation video or article.

[0072] For example, reference Figure 4When the target operation behavior includes a click operation behavior and an order operation behavior, the screening and determining process of the target operation object information may include the following steps.

[0073] A: From the user behavior data, filter out the click object information corresponding to the click operation behavior within the target time range.

[0074] B: Filter out the order object information corresponding to the order operation behavior from the user behavior data.

[0075] C: Use the click object information and order object information as the target operation object information.

[0076] In this example implementation, the target time range is a preset time range (such as 5-10 seconds) adjacent to the search time point, and the search time point is the time point when the user searches for the trigger word. When the application scenario is an electronic mall, the target operation behavior can also include an add-to-cart operation behavior, and the add-to-cart object information can be filtered out from the user behavior data. The click object information, the add-to-cart object, and the order object information can respectively include one or more of the object's name, logo, code, content, and category. The object can be a product in an electronic mall, a video on an information recommendation platform, etc. This example does not limit this.

[0077] In step S240, the target operation object information is pushed to the application platform.

[0078] The method disclosed in the present invention can be performed offline or online. When the information recommendation process is performed offline, the target operation object information determined offline can be stored in a database, and the application platform obtains the push information by calling the database.

[0079] In some embodiments, for hot events with short duration and low frequency (such as Chinese Valentine's Day, Mid-Autumn Festival, etc.), the target operation object information is obtained based on the user's behavior within the most recent holiday time period (i.e., one year ago). The target operation object information will hardly contain currently listed objects, that is, the recommended information lacks newly listed objects. The present disclosure compensates for this problem by adding the following steps S510-S530 before pushing the information.

[0080] Step S510: vectorize the target operation object information.

[0081] In this example implementation, the target operation object information may include target objects and corresponding target categories arranged in chronological order of the operation behavior. Exemplarily, the trained word vector generation model may be used to vectorize the target objects and corresponding target categories. The word vector generation model may be any one of a continuous skip-gram model (Skip-Gram), a continuous bag-of-words model (CBOW), a subword information skip-gram model (SISG), and an enhanced graph embedding model (EGES).

[0082] For example, for a trained SISG, when a target object sequence is input, the model outputs a vectorized representation of each target object (i.e., the context of the target object). Similarly, when a target category sequence is input, the model outputs a vectorized representation of each target category (i.e., the context of the target category).

[0083] Step S520: Based on the vectorized representation result, similar vectors are searched in a pre-established vector search library to obtain vector representations of similar objects.

[0084] In this example implementation, the similarity between the vectorized representation result and each vector in the vector search library can be calculated, and the vectors with a similarity greater than a specified threshold (90%) can be used as similar object vector representations. Vectors with similarities in the first specified number of digits can also be selected as similar object vector representations, and other search matching rules can also be set, which are not limited in this example.

[0085] For example, reference Figure 6 , you can build a vector retrieval library by following the steps below.

[0086] Get the candidate object dataset based on the click data and order data within the specified time range.

[0087] In this example implementation, the specified time range can be set according to the actual situation. For example, historical data of a certain platform within a few days (7-10 days) from the current time can be selected. Click data refers to object information with a click count greater than a specified number, and order data refers to object information that generates an order operation. For example, a product sku set consisting of products with a click count greater than 2 within 7 days on a certain application platform and products that are available within 7 days can be selected as a candidate object data set.

[0088] The trained word vector generation model is used to vectorize the candidate object dataset to obtain the candidate object vector representation results.

[0089] In this example implementation, the object sequence in the candidate object data set may be input into a trained word vector generation model to output the context of the object sequence (ie, the candidate object vector representation result).

[0090] Exemplarily, the training process of the word vector generation model is as follows:

[0091] Get the operation object sequence data set within the specified time range as training sample data; one operation object sequence is regarded as a training sample.

[0092] In this example, an operation object sequence can be an operation object sequence formed by continuous operations after the user enters the search term. Continuous operations refer to operations of the same user whose operation behavior interval is less than a set threshold (such as 3 days). The specified time range can be set according to the data volume and purpose of the application platform, such as one month from the current time. The operation object sequence can select highly active operation objects, such as objects that have been viewed a large number of times or have placed orders within a certain period of time.

[0093] In this example, an operation object in a training sample is used as an input word, and the adjacent operation objects of the operation object are used as output word labels. For each object in the operation object sequence, the objects within a certain distance range adjacent to it can be considered as the context of the object.

[0094] Input each training sample into the word vector generation model to obtain the word vector of the corresponding output word.

[0095] In this example, the word vector for the output word is the context of the input word predicted by the model.

[0096] With the goal of maximizing the log-likelihood function between the input word and the corresponding output word, the parameters of the word vector generation model are updated to train the word vector generation model.

[0097] In this example implementation, a training cutoff condition may be set, such as when the target converges, the training is complete.

[0098] For example, taking the word vector generation model SISG as an example, given a vocabulary of size w, each input word is represented as w according to its index i (i) . Let the neighboring current input word w (t) The two words of are its output words (i.e. context), then use w (t) Predict the context word w (t-2) ~w (t+2) The log-likelihood function of is expressed as:

[0099]

[0100] Where T represents the length of the input sequence, w (c) Represents the input word w (t) The context of C t Indicates the window size for determining the context of t, c represents the index number of the output word, t is the index number of the input word, and w (c) |w (t) Indicates that when the input word w (t) When the output word w (c) probability.

[0101] Based on the vector representation results of the candidate objects, a vector retrieval library is established.

[0102] In this example implementation, the candidate object vector representation results may include vector representation results of the candidate object and its category, and the candidate object vector representation results are used to establish a Faiss library, that is, a vector retrieval library.

[0103] Step S530: Use the similar object information corresponding to the similar object vector representation as push information.

[0104] In this example implementation, similar object information can be determined based on similar object vector representations. Through this example, push information can be expanded to include objects of recent new merchants (such as products or videos), thereby increasing the sense of surprise brought by the push information to users and thus improving the conversion rate of the push information.

[0105] In some embodiments, the target operation object information includes a target item and its target category. Before vectorizing the target operation object information, the method further includes:

[0106] Identify the dirty data categories.

[0107] In this example, dirty data category refers to data that has a negative impact on the results, such as user misoperation or malicious order brushing. Dirty data category can be defined as an object category that indicates the object click frequency is greater than the preset second threshold (such as more than 10 times in one minute) (malicious order brushing) or the object detail page stay time is less than the preset third threshold (such as 2 seconds) (misoperation).

[0108] Filter out the object information corresponding to dirty data categories and preset sensitive categories and long-tail categories in the target operation object information; filter out the delisted object information in the target operation object information.

[0109] In this example implementation, sensitive categories refer to categories of objects that may cause user complaints, and long-tail categories refer to niche categories that are less relevant to hot topics. For example, for the Mid-Autumn Festival hot topic, the book category can be considered a long-tail category. Sensitive categories and long-tail categories can be configured in advance.

[0110] Exemplarily, the number of corresponding operation behaviors can be counted through the click behavior, ordering behavior, etc. of the objects corresponding to the category, and the objects with a certain proportion (such as 20%) of the operation quantity can be selected, and the remaining categories can be filtered out, so as to filter out the long-tail categories.

[0111] In some embodiments, the method further includes: clustering the trigger words to obtain a plurality of scene categories.

[0112] In this example implementation, trigger words can be clustered according to preset rules, for example, multiple scene categories and scene words under each scene category are pre-configured, the similarity between scene words and trigger words is calculated, and trigger words with high similarity to scene words under the same scene category are clustered into one category. Different trigger words can also be clustered using existing clustering methods (such as K-means, Gaussian mixture models GMMs), which is not limited in this example. For example, if the trigger words are short sleeves, shorts, and skirts, they can be clustered into clothing scenes.

[0113] Acquiring training sample data includes: acquiring, according to multiple scene categories, an operation object sequence data set under each scene category as training sample data.

[0114] In this example implementation, the scene category words obtained by clustering can be used to obtain training sample data for a word vector generation model to ensure the diversity of training samples and the accuracy of the trained word vector generation model.

[0115] For example, the implementation process of the information push method disclosed in the present invention is as follows: Figure 7 As shown in the figure, the process involves the Bert model, TextRank model, word vector generation model and vector retrieval library, etc. The push information is determined through the above models and retrieval library and pushed to the application platform corresponding to the user terminal. It can be implemented by the following steps.

[0116] The first step is to capture the target event, and then proceed to the next step.

[0117] In this example, the target event may be a current hot event.

[0118] The second step is to obtain user behavior data of the application platform. User behavior data can be historical behavior data within a certain time range from the current time, such as data within 7 days from the current time.

[0119] The third step is to filter out the target search terms associated with the target event from the user behavior data.

[0120] The fourth step is to input the target search term into the Bert model to extract similar semantic features and obtain similar search terms.

[0121] In the fifth step, similar search terms and target search terms are input into the TextRank model for keyword extraction. Based on the extracted target keywords and similar search terms, the trigger words corresponding to the target event are determined.

[0122] The sixth step is to obtain the target operation behavior after the user searches for the trigger word within the target period, and filter out the target operation object information based on the target operation behavior.

[0123] The seventh step is to use the trained word vector generation model to vectorize the target operation object information.

[0124] In the eighth step, based on the vectorized representation results, similar vectors are searched in the pre-established vector retrieval library to obtain vector representations of similar objects.

[0125] In this example, the trained word vector generation model can be used to build a vector retrieval library.

[0126] In the ninth step, the similar object information and the target operation object information corresponding to the similar object vector representation are pushed to the application platform of the user terminal.

[0127] The push information determination process in the above embodiments can be implemented offline or online. When implemented offline, the data involved in the process can be offline data, which is mainly stored in HDFS (Hadoop Distributed File System) or in Redis storage. The offline data is processed and read by Hadoop and Hive environment. After the processing is completed, a Faiss library will be established based on offline hotspot objects or new objects. Hadoop is a framework for running applications on large clusters built with general-purpose hardware. Hive is a data warehouse tool based on Hadoop that analyzes and manages data stored in HDFS.

[0128] Experimental verification

[0129] The disclosed method is used to push information to an e-commerce platform. Taking the Chinese Valentine's Day hot spot as an example, the pushed information under this hot spot contains a total of more than 30,000 items, more than 60 categories of goods, and the proportion of recommended new products is close to 20%. This shows that after the disclosed method is applied to the e-commerce platform, the pushed information contains rich categories of goods with high novelty, thereby improving the user experience.

[0130] The above push information is pushed to the e-commerce platform, that is, the hot circle product results are actually recommended and delivered. The platform's browsing depth, recommended uctr (uctr = users with clicking behavior / total number of visiting users), average exposure to novel categories per capita and average clicks on novel categories per capita are all improved to varying degrees, indicating that the push information determined by the present invention has a strong novelty, can enhance users' interest in the push information, and thus improve the conversion rate of the push information.

[0131] In addition, the click-through rate and conversion rate of the e-commerce platform's recommended information to users using the method disclosed herein within seven days of the Chinese Valentine's Day were monitored together with the e-commerce platform's total click-through rate (number of clicks / number of exposures) and conversion rate (number of orders / number of clicks). The results showed that the recommended information disclosed herein could increase the click-through rate by more than 0.2% and the conversion rate by more than 1%, indicating that the method disclosed herein can significantly improve users' click-through rate and conversion rate.

[0132] The present disclosure is aimed at the problem that the existing object recommendation completely relies on the user's historical behavior data, resulting in low click-through rate and conversion rate of the recommended information, and failing to bring a sense of surprise to the user. On the one hand, the present disclosure captures the hot events of the current period and generates recommendation information based on the hot events to improve the novelty of the recommended information and improve the click-through rate and conversion rate. On the other hand, for hot events with short duration and long cycle intervals, the present disclosure establishes a vector retrieval library containing recent new product information through a word vector generation model, taking into account the lack of new object information that has been recently put on the shelves in the push information obtained based on the data of the previous hot event cycle, thereby expanding the categories of recommended objects through retrieval in the vector retrieval library, adding new object information on the shelves, ensuring the novelty of the recommended information, and further improving the conversion rate of the pushed information. In addition, by extracting similar semantic features of the target search terms and expanding the trigger words of the hot events, the object coverage and category richness of the pushed information are guaranteed.

[0133] Furthermore, in this exemplary embodiment, an information push device 800 is also provided. The information push device 800 can be applied to a terminal or a server. Figure 8As shown, the information push device 800 may include: an acquisition module 810, an extraction module 820, a determination module 830 and a push module 840, wherein the acquisition module 810 is used to acquire user behavior data of the application platform after determining that a target event has occurred on the application platform; the extraction module 820 is used to screen out target search terms associated with the target event from the user behavior data; perform keyword extraction on the target search terms to obtain trigger words corresponding to the target event; the determination module 830 is used to acquire the target operation behavior after the user searches for the trigger word within the target period, and determine the target operation object information based on the target operation behavior; and the push module 840 is used to push the target operation object information to the application platform as push information.

[0134] In an exemplary embodiment of the present disclosure, the device 800 also includes a monitoring module, which is used to monitor user behavior data of the application platform, the user behavior data including a search behavior growth rate; in response to the search behavior growth rate for a candidate search term being greater than a preset first threshold, it is determined that a target event has occurred.

[0135] In an exemplary embodiment of the present disclosure, the device 800 also includes a retrieval module, which is used to vectorize the target operation object information before pushing the target operation object information; based on the vectorization representation result, perform a similar vector search in a pre-established vector retrieval library to obtain a similar object vector representation; and use the similar object information corresponding to the similar object vector representation as the push information.

[0136] In an exemplary embodiment of the present disclosure, the device 800 also includes a word expansion module, which is used to vectorize the target search word before performing keyword extraction on the target search word to obtain a target search word vector; and extract similar semantic features from the target search word vector to obtain similar search words.

[0137] In an exemplary embodiment of the present disclosure, the extraction module 820 can also be used to use similar search terms and target search terms as sentences to be extracted; perform keyword extraction on the sentences to be extracted to obtain target keywords; and use the target keywords as trigger words corresponding to the target events.

[0138] In an exemplary embodiment of the present disclosure, the user behavior data includes user search terms, and the extraction module also includes a screening submodule, which can be used to: configure an associated word mapping table of the target event; and screen out the target search terms from the user search terms according to the associated word mapping table.

[0139] In an exemplary embodiment of the present disclosure, the target operation behavior includes a click operation behavior and an order operation behavior, and the determination module 830 can also be used to: filter out the click object information corresponding to the click operation behavior within the target time range from the user behavior data, the target time range is a preset time range adjacent to the search time point, and the search time point is the time point when the user searches for the trigger word; filter out the order object information corresponding to the order operation behavior from the user behavior data; and use the click object information and the order object information as the target operation object information.

[0140] In an exemplary embodiment of the present disclosure, the target operation object information includes target objects and corresponding target categories arranged in chronological order of operation behaviors. The retrieval module can also be used to: use the trained word vector generation model to vectorize the target objects and corresponding target categories respectively.

[0141] In an exemplary embodiment of the present disclosure, the device 800 also includes a training module, which can be used to: obtain an operation object sequence data set within a specified time range as training sample data; an operation object sequence as a training sample; an operation object in a training sample as an input word, and the adjacent operation objects of the operation object as output word labels; input each training sample into a word vector generation model to obtain a word vector of the corresponding output word; with the goal of maximizing the log-likelihood function between the input word and the corresponding output word, update the parameters of the word vector generation model to train the word vector generation model.

[0142] In an exemplary embodiment of the present disclosure, the device 800 also includes a retrieval library construction module, which can be used to obtain a candidate object data set based on click data and order data within a specified time range; use the trained word vector generation model to vectorize the candidate object data set to obtain a candidate object vector representation result; and establish a vector retrieval library based on the candidate object vector representation result.

[0143] In an exemplary embodiment of the present disclosure, the apparatus 800 further includes a clustering module, which can be used to cluster trigger words to obtain multiple scene categories.

[0144] The training module can also be used to obtain an operation object sequence data set under each scene category as training sample data according to multiple scene categories.

[0145] In an exemplary embodiment of the present disclosure, the target operation object information includes the target item and its target category, and the device 800 also includes a filtering module, which can be used to: determine the dirty data category before vectorizing the target operation object information; the dirty data category is used to indicate the object category whose object click frequency is greater than a preset second threshold or the object category whose residence time on the object details page is less than a preset third threshold; filter out the object information corresponding to the dirty data category and the preset sensitive category and long-tail category in the target operation object information; filter out the delisted object information in the target operation object information.

[0146] The specific details of each module or unit in the above-mentioned information pushing device have been described in detail in the corresponding information pushing method, so they will not be repeated here.

[0147] As another aspect, the present application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiment; or may exist independently without being assembled into the electronic device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device implements the method in the following embodiment. For example, the electronic device may implement Figure 2 to Figure 7 The various steps shown, etc.

[0148] It should be noted that the computer-readable medium shown in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0149] Refer to the following Fig. 9 hereinafter describes an electronic device 900 according to such an embodiment of the present disclosure. Fig. 9 The electronic device 900 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0150] like Fig. 9 As shown, the electronic device 900 is in the form of a general computing device. The components of the electronic device 900 may include, but are not limited to: the at least one processing unit 910, the at least one storage unit 920, a bus 930 connecting different system components (including the storage unit 920 and the processing unit 910), and a display unit 940.

[0151] The storage unit stores program codes, which can be executed by the processing unit 910, so that the processing unit 910 executes the steps according to various exemplary embodiments of the present disclosure described in the above “Exemplary Method” section of this specification.

[0152] The storage unit 920 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 9201 and / or a cache storage unit 9202 , and may further include a read-only storage unit (ROM) 9203 .

[0153] The storage unit 920 may also include a program / utility 9204 having a set (at least one) of program modules 9205, such program modules 9205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0154] Bus 930 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0155] The electronic device 900 may also communicate with one or more external devices 970 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 900, and / or communicate with any device that enables the electronic device 900 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 950. Furthermore, the electronic device 900 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 960. As shown, the network adapter 960 communicates with other modules of the electronic device 900 via a bus 930. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 900, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RA identification systems, tape drives, and data backup storage systems.

[0156] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the embodiment of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the embodiment of the present disclosure.

[0157] In addition, the above-mentioned figures are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present disclosure, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0158] It should be noted that, although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in the specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc., which shall all be considered as part of the present disclosure.

[0159] It should be understood that the disclosure disclosed and defined in this specification extends to all alternative combinations of two or more individual features mentioned or evident in the text and / or the accompanying drawings. All these different combinations constitute multiple alternative aspects of the disclosure. The embodiments of this specification illustrate the best mode known for implementing the disclosure and will enable those skilled in the art to utilize the disclosure.

Claims

1. A method for pushing information, It is characterized in that include: After determining that a target event occurs on the application platform, obtaining user behavior data of the application platform; Filtering the target search terms associated with the target event from the user behavior data; Perform keyword extraction on the target search term to determine the trigger word corresponding to the target event; Obtaining a target operation behavior after the user searches for the trigger word within a target period, and determining target operation object information according to the target operation behavior; The target operation object information is pushed to the application platform.

2. The information push method according to claim 1, It is characterized in that The method further comprises: Monitoring user behavior data of the application platform, wherein the user behavior data includes a search behavior growth rate; In response to the search behavior growth rate for the candidate search term being greater than a preset first threshold, it is determined that the target event occurs.

3. The information push method according to claim 1, It is characterized in that Before pushing the target operation object information, the method further includes: Vectorizing the target operation object information; Based on the vectorized representation results, similar vectors are searched in the pre-established vector search library to obtain vector representations of similar objects; The similar object information corresponding to the similar object vector representation is used as push information.

4. The information push method according to claim 1, It is characterized in that Before extracting keywords from the target search term, the method further includes: Performing vectorization processing on the target search term to obtain a target search term vector; Similar semantic features are extracted from the target search word vector to obtain similar search words.

5. The information push method according to claim 4, It is characterized in that The extracting keywords from the target search term includes: Taking the similar search words and the target search words as sentences to be extracted; Extracting keywords from the sentence to be extracted to obtain target keywords; The similar search terms and the target keywords are used as trigger words corresponding to the target event.

6. The information push method according to claim 1, It is characterized in that The user behavior data includes user search terms, and screening out target search terms associated with the target event from the user behavior data includes: Configuring a mapping table of associated words for the target event; The target search term is screened out from the user search terms according to the associated term mapping table.

7. The information push method according to claim 1, It is characterized in that The target operation behavior includes a click operation behavior and an order operation behavior, and determining the target operation object information according to the target operation behavior includes: Filtering out click object information corresponding to click operation behaviors within a target time range from the user behavior data, wherein the target time range is a preset time range adjacent to a search time point, and the search time point is a time point when the user searches for the trigger word; Filtering out order object information corresponding to the order operation behavior from the user behavior data; The click object information and the order object information are used as target operation object information.

8. The information push method according to claim 3, It is characterized in that The target operation object information includes target objects and corresponding target categories arranged in chronological order of operation behaviors, and vectorizing the target operation object information includes: The trained word vector generation model is used to vectorize and represent the target object and the corresponding target category respectively.

9. The information push method according to claim 8, It is characterized in that The method further comprises: Obtain an operation object sequence data set within a specified time range as training sample data; an operation object sequence is used as a training sample; an operation object in a training sample is used as an input word, and the adjacent operation objects of the operation object are used as output word labels; Input each of the training samples into the word vector generation model to obtain the word vector of the corresponding output word; With the goal of maximizing the log-likelihood function between the input word and the corresponding output word, the parameters of the word vector generation model are updated to train the word vector generation model.

10. The information push method according to claim 3, It is characterized in that The method further comprises: Obtain candidate object datasets based on click data and order data within a specified time range; Using the trained word vector generation model, the candidate object dataset is vectorized to obtain a candidate object vector representation result; A vector retrieval library is established based on the candidate object vector representation results.

11. The information push method according to claim 9, It is characterized in that The method further comprises: Clustering the trigger words to obtain multiple scene categories; The obtaining of training sample data comprises: According to the multiple scene categories, an operation object sequence data set under each scene category is obtained as training sample data.

12. The information push method according to claim 3, It is characterized in that The target operation object information includes a target item and its target category. Before vectorizing the target operation object information, the method further includes: Determine a dirty data category; the dirty data category is used to indicate an object category whose object click frequency is greater than a preset second threshold or an object category whose dwell time on the object detail page is less than a preset third threshold; Filtering out object information corresponding to the dirty data category and preset sensitive categories and long-tail categories from the target operation object information; Filter out the delisted object information in the target operation object information.

13. An information push device, It is characterized in that The device comprises: An acquisition module, used to acquire user behavior data of the application platform after determining that a target event occurs on the application platform; An extraction module is used to filter out target search words associated with the target event from the user behavior data; perform keyword extraction on the target search words to obtain trigger words corresponding to the target event; A determination module, used to obtain a target operation behavior after the user searches for the trigger word within a target period, and determine target operation object information according to the target operation behavior; The push module is used to push the target operation object information as push information to the application platform.

14. An electronic device, It is characterized in that include: one or more processors; as well as A storage device, used for storing one or more programs, when the one or more programs are executed by the one or more processors, enables the one or more processors to implement the method according to any one of claims 1 to 11.

15. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the program is executed by a processor, the method according to any one of claims 1 to 11 is implemented.

Citation Information

Patent Citations

  • Information pushing method and device

    CN105488112A

  • Method and device for providing search suggestion words on basis of activity states

    CN107368553A

  • Information recommendation method and device

    CN108346075A

  • Keyword pushing method, apparatus and device

    CN108664513A

  • Information search method and device, medium and apparatus

    CN110609956A