Label determination method, apparatus and electronic device

Abstract

The application provides a label determination method, device and electronic equipment, and the method comprises the following steps: acquiring an identification sequence of resources concerned by a target user within a predetermined time length; determining the correlation between adjacent resource identifications in the identification sequence, and determining a first vector matrix for representing the identification sequence according to the correlation between the adjacent resource identifications; acquiring a URL set formed by page element URLs respectively associated with each resource attention behavior of the target user within the predetermined time length, wherein the page element URLs associated with each resource attention behavior contain at least one resource identification; determining a second vector matrix corresponding to the resources concerned by the target user within the predetermined time length according to the URL set; and fusing the first vector matrix and the second vector matrix to obtain a label determined for the target user. According to the user behavior sequence and the page element URLs, the content concerned by the user is determined, and then the user label is determined, so that the determined label is more accurate.

Description

Technical Field

[0001] This application relates to the field of automation technology, and in particular to label identification methods, apparatus and electronic equipment. Background Technology

[0002] Currently, user tags are typically determined based on the number of clicks a user makes on a target resource over a period of time. However, if a user's click behavior is accidental, the click count can increase due to the accident, thus affecting the click count statistics and consequently the tag determination results.

[0003] For example, the number of clicks a user makes on mother and baby products within a week can be used to determine whether to label the user as a "mom". If a user receives an ad pop-up or push notification and accidentally opens the ad instead of closing it, this accidental action is considered as the user being interested in the ad.

[0004] This shows that the accuracy of existing label determination methods is not high. Summary of the Invention

[0005] The purpose of this application is to provide a tag identification method, apparatus, and electronic device to solve the problems of "low task completion efficiency and resource waste".

[0006] To address the aforementioned technical problems, this specification provides a tag determination method, comprising: acquiring a sequence of identifiers of resources followed by a target user within a predetermined time period; determining the association relationship between adjacent resource identifiers in the identifier sequence, and determining a first vector matrix for representing the identifier sequence based on the association relationship between adjacent resource identifiers; acquiring a set of URLs formed by the URLs of page elements associated with each resource follow-up behavior of the target user within the predetermined time period, wherein the URL of the page element associated with each resource follow-up behavior contains at least one resource identifier; determining a second vector matrix corresponding to the resources followed by the target user within the predetermined time period based on the set of URLs; and fusing the first vector matrix and the second vector matrix to obtain tags determined for the target user.

[0007] In some embodiments, obtaining a sequence of identifiers of resources that a target user is interested in within a predetermined time period includes: obtaining the identifiers of each resource browsed by the target user in a resource platform within a predetermined time period; wherein the resources in the resource platform are organized by category, and the name or category of the resource is used as the identifier of the resource; and arranging the identifiers of each resource in the order of browsing to obtain a sequence of resource identifiers.

[0008] In some embodiments, the identifiers of various resources are arranged in the order of browsing to obtain a resource identifier sequence, including: obtaining the identifiers of resources searched by the target user in the resource platform within a predetermined time period, and the identifiers of various resources followed by the target user after the search; wherein, the resources in the resource platform are organized by category, and the name or category of the resource is used as the resource identifier; the identifiers of the searched resources and the identifiers of the browsed resources are sorted in chronological order to obtain a resource identifier sequence.

[0009] In some embodiments, determining a second vector matrix associated with the target resources of interest to the target user based on the URL set includes: performing the following operations on each URL in the URL set: dividing the current URL into multiple segments according to a preset identifier of the current URL; determining the vector corresponding to each segment; determining the vector matrix corresponding to the current URL based on the vector corresponding to each segment; performing a convolution operation on the vector matrix corresponding to the current URL to obtain the feature value vector corresponding to the current URL; determining the feature value vector matrix corresponding to the URL set based on the feature value vectors corresponding to multiple URLs; and using the feature value vector matrix as the second vector matrix.

[0010] In some embodiments, performing a convolution operation on the vector matrix corresponding to the current URL to obtain the feature value vector corresponding to the current URL includes: setting a sliding window with the same dimension as a preset convolution kernel to slide on the vector matrix corresponding to the current URL; calculating the feature values ​​corresponding to each position when the sliding window slides to a preset step size in the following manner: performing an inner product operation between the multiple data enclosed by the sliding window when it slides to the current position on the vector matrix corresponding to the current URL and the data in the convolution kernel to obtain the feature values; and determining the feature value vector based on the feature values ​​corresponding to the multiple positions of the sliding window during the sliding process.

[0011] In some embodiments, fusing the first vector matrix and the second vector matrix to obtain a tag determined for the target user includes: obtaining a set of resources that the target user has been following within the predetermined time period and statistical data on the following behaviors corresponding to the resources; fusing the first vector matrix, the second vector matrix and the statistical data to obtain the user's tag.

[0012] In some embodiments, the statistical data of the attention behavior corresponding to the resource includes at least one of the following: the number of clicks on the resource, the browsing time of the resource, the number of times the resource was re-viewed, the number of times the resource was favorited, the number of times the resource was reposted, and the number of times the resource was liked.

[0013] In some embodiments, obtaining tags determined for a target user includes: determining primary tags and secondary tags for the target user, wherein the primary tags include multiple secondary tags.

[0014] A second aspect of this specification provides a tag determination method, comprising: acquiring a sequence of identifiers of resources of interest to a target user within a predetermined time period; determining the association between adjacent resource identifiers in the identifier sequence, and determining a first vector matrix for characterizing the identifier sequence based on the association between adjacent resource identifiers; and determining the user's tag based on the first vector matrix.

[0015] A third aspect of this specification provides a tag determination method, comprising: obtaining a URL set formed by URLs of page elements associated with each resource-focusing behavior of a target user within a predetermined time period, wherein the URL of the page element associated with each resource-focusing behavior contains at least one resource identifier; determining a second vector matrix corresponding to the resources followed by the target user within the predetermined time period based on the URL set; and fusing the second vector matrix to obtain a tag determined for the target user.

[0016] A fourth aspect of this specification provides a tag determination apparatus, comprising: a first acquisition unit, configured to acquire a sequence of identifiers of resources followed by a target user within a predetermined time period; a first determination unit, configured to determine the association relationship between adjacent resource identifiers in the identifier sequence, and determine a first vector matrix for characterizing the identifier sequence based on the association relationship between adjacent resource identifiers; a second acquisition unit, configured to acquire a set of URLs formed by the URLs of page elements associated with each resource follow-up behavior of the target user within the predetermined time period, wherein the URL of the page element associated with each resource follow-up behavior contains at least one resource identifier; a second determination unit, configured to determine a second vector matrix corresponding to the resources followed by the target user within the predetermined time period based on the set of URLs; and a fusion unit, configured to fuse the first vector matrix and the second vector matrix to obtain tags determined for the target user.

[0017] In some embodiments, the first acquisition unit includes: a first acquisition subunit, configured to acquire the identifiers of various resources browsed by the target user in the resource platform within a predetermined time period; wherein the resources in the resource platform are organized according to categories, and the names or categories of the resources are used as resource identifiers; and a first sorting subunit, configured to arrange the identifiers of each resource according to the order of browsing to obtain a sequence of resource identifiers.

[0018] In some embodiments, the first sorting subunit includes: a second acquisition subunit, configured to acquire the identifiers of resources searched by the target user in the resource platform within a predetermined time period, and the identifiers of each resource that the target user is interested in after the search; wherein, the resources in the resource platform are organized according to categories, and the names or categories of the resources are used as resource identifiers; the second sorting subunit is configured to sort the identifiers of the searched resources and the identifiers of each browsed resource according to time sequence to obtain a resource identifier sequence.

[0019] In some embodiments, the second determining unit includes: a first determining subunit, configured to perform the following operations on each URL in the URL set: dividing the current URL into multiple segments according to a preset identifier of the current URL; determining the vector corresponding to each segment respectively; determining the vector matrix corresponding to the current URL according to the vector corresponding to each segment; performing a convolution operation on the vector matrix corresponding to the current URL to obtain the feature value vector corresponding to the current URL; a second determining subunit, configured to determine the feature value vector matrix corresponding to the URL set according to the feature value vectors corresponding to multiple URLs; and a third determining subunit, configured to use the feature value vector matrix as a second vector matrix.

[0020] In some embodiments, the first determining subunit performs a convolution operation on the vector matrix corresponding to the current URL in the following manner to obtain the feature vector corresponding to the current URL: a sliding window with the same dimension as the preset convolution kernel is set to slide on the vector matrix corresponding to the current URL; the feature values ​​corresponding to the sliding window at each position with a preset step size are calculated in the following manner: the feature values ​​are obtained by performing an inner product operation between the multiple data enclosed by the sliding window at the current position on the vector matrix corresponding to the current URL and the data in the convolution kernel; the feature vector is determined based on the feature values ​​corresponding to the multiple positions of the sliding window during the sliding process.

[0021] In some embodiments, the fusion unit includes: a third acquisition subunit, configured to acquire a set of resources followed by the target user within the predetermined time period and statistical data on the following behaviors corresponding to the resources; and a fusion subunit, configured to fuse the first vector matrix, the second vector matrix, and the statistical data to obtain the user's tags.

[0022] In some embodiments, the statistical data of the attention behavior corresponding to the resource includes at least one of the following: the number of clicks on the resource, the browsing time of the resource, the number of times the resource was re-viewed, the number of times the resource was favorited, the number of times the resource was reposted, and the number of times the resource was liked.

[0023] In some embodiments, the target user's tags include primary tags and secondary tags, wherein the primary tags include multiple secondary tags.

[0024] A fifth aspect of this specification provides a tag determination apparatus, comprising: a first acquisition unit, configured to acquire a sequence of identifiers of resources of interest to a target user within a predetermined time period; a second determination unit, configured to determine the association relationship between adjacent resource identifiers in the identifier sequence, and determine a first vector matrix for characterizing the identifier sequence based on the association relationship between adjacent resource identifiers; and a third determination unit, configured to determine a user's tag based on the first vector matrix.

[0025] A sixth aspect of this specification provides a tag determination apparatus, comprising: a second acquisition unit, configured to acquire a URL set formed by URLs of page elements associated with each resource-focusing behavior of a target user within a predetermined time period, wherein the URL of the page element associated with each resource-focusing behavior contains at least one resource identifier; a second determination unit, configured to determine a second vector matrix corresponding to the resources followed by the target user within the predetermined time period based on the URL set; and a fourth determination unit, configured to determine the tags of the target user based on the second vector matrix.

[0026] A seventh aspect of this specification provides an electronic device, comprising: a memory and a processor, wherein the processor and the memory are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to implement the steps of the method described in any one of the first to third aspects.

[0027] An eighth aspect of this specification provides a computer storage medium storing computer program instructions that, when executed by a processor, implement the steps of the method described in any one of the first to third aspects.

[0028] A ninth aspect of this specification provides a computer program product comprising a computer program that, when executed by a processor, implements the steps of the method described in any one of the first to third aspects.

[0029] The tag determination method, apparatus, and electronic device provided in this specification use a sequence of identifiers for resources that a target user is interested in within a predetermined time period to represent the target user's behavioral sequence during that period. Then, based on the association between adjacent resource identifiers in the identifier sequence, a first vector matrix is ​​determined that can be used to identify the identifier sequence. This first vector matrix can be considered as a summary of the series of behaviors represented by the identifier sequence. Tags for the target user determined based on this summary information are less affected by erroneous operations, thus the determined tags are more accurate. Furthermore, a set of URLs is formed using the URLs of page elements associated with each resource-interesting behavior of the target user within the predetermined time period. Based on URL naming rules, resource information is extracted from each URL in the URL set, and the target user's tags are determined based on the extracted resource information. In general, the same resource identifier X may exist under both category A and category B, and categories A and B may be completely different categories. Therefore, determining the content that the user is interested in solely based on resource identifier X is inaccurate. The above-described method for determining tags extracts the content that users are interested in from the URLs of page elements. Since the URL naming contains resource category information, the above method can more accurately determine the content that users are interested in, and thus the tags determined based on that content are more accurate.

[0030] The tag determination method, apparatus, and electronic device provided in this specification determine user tags based on the identifier sequence of resources followed by the target user within a predetermined time period, the URL set formed by the URLs of page elements associated with each resource follow-up behavior of the target user within the predetermined time period, the set of resources followed by the target user within the predetermined time period, and the statistical data of the follow-up behavior corresponding to the resources. These three factors are comprehensively considered to determine user tags. The data sources vary greatly, which can make up for the defects of each data source itself, and thus the tags of the target users determined based on these data sources are more accurate. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0032] Figure 1 This specification illustrates a network model used in one embodiment of the label determination method provided herein;

[0033] Figure 2 This diagram illustrates the URL segmentation and the corresponding numbering and vectors for each segment.

[0034] Figure 3 A flowchart of the label determination method provided in this specification is shown;

[0035] Figure 4 A flowchart of a method for obtaining an identifier sequence is shown;

[0036] Figure 5 A flowchart of another method for obtaining the identifier sequence is shown;

[0037] Figure 6 This diagram illustrates a flowchart of performing a convolution operation on the vector matrix corresponding to a URL.

[0038] Figure 7 Another flowchart of the label determination method provided in this specification is shown;

[0039] Figure 8 This document shows yet another flowchart of the label determination method provided in this specification;

[0040] Figure 9 This document shows yet another flowchart of the label determination method provided in this specification;

[0041] Figure 10 This document shows yet another flowchart of the label determination method provided in this specification;

[0042] Figure 11 A schematic block diagram of the label determining device provided in this specification is shown;

[0043] Figure 12 Another schematic diagram of the label determining device provided in this specification is shown;

[0044] Figure 13 This specification shows yet another block diagram of the label determining device provided in this manual;

[0045] Figure 14 A schematic block diagram of the electronic device provided in this specification is shown. Detailed Implementation

[0046] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0047] This manual provides a label identification method, employing methods such as... Figure 1 The network model shown is as follows. Figure 1 As shown, the network model mainly consists of four sub-networks: BERT network, convolutional neural network, feedforward neural network, and a sub-network with fully connected layers as the main body. The network model as a whole has three inputs and one output, where each input corresponds to the output of one of the sub-networks of BERT network, convolutional neural network, and feedforward neural network, and the output is the label determined by the user.

[0048] The input data for the BERT network consists of identifiers of various resources that a user has been following over a period of time. These resource identifiers are sorted according to the time when the resource was followed. The identifiers of the resources that a user has been following can be identifiers of the resource category, identifiers of the resource itself, or even search terms entered by the user.

[0049] For example, in a mobile banking app, a user searches for "wealth management," then clicks "premium wealth management," enters the premium wealth management channel, and subsequently clicks on "XX stable wealth management product" and "XX open-ended private banking wealth management product." This generates a sequence of resource identifiers: "[CLS]wealth management[SEP]premium wealth management[SEP]XX stable wealth management product[SEP]XX open-ended private banking wealth management product[SEP]...". This creates a text sequence of resource identifiers clicked by the user. [CLS] and [SEP] are special tokens required as input to the BERT model. [SEP] can be understood as a separator between two adjacent clicked resource identifiers; the content between two [SEP]s represents the identifier of a resource clicked by the user. [CLS] indicates that the category or content of the resource identifier sequence following [CLS] needs to be output.

[0050] BERT (Bidirectional Encoder Representation from Transformers) is a pre-trained language representation model. It uses MLM and NSP to pre-train bidirectional Transformers to generate deep bidirectional language representations. Here, 6L indicates a 6-layer transformer encoding network structure. The output of the BERT model is a multi-dimensional vector representing a sequence of user actions.

[0051] The input to a convolutional neural network is a set of URLs of page elements associated with different resource attention behaviors over a period of time. Page elements can be images, videos, etc. on the page.

[0052] These URLs are named according to certain rules, which require that the URL contain at least one resource identifier. Because the URLs contain resource identifiers, the resources the user is interested in can be parsed from the URLs of page elements associated with the user's resource interest behavior. For example, if a user clicks on a picture of a necklace on a page, and the URL of that picture is http: / / xxx.xxx.com / item / jewellery / ctf / 5673425.html, then "jewellery" (representing jewelry) and "ctf" (representing Chow Tai Fook, a jewelry brand) are resource identifiers. Therefore, the resource identifiers in the URL reveal that the user is interested in jewelry.

[0053] The URLs in the URL set are first preprocessed as follows before being input into the convolutional neural network: the URLs are divided into multiple segments based on special identifiers (such as " / ") in the URLs, each segment is encoded to obtain an encoding ID, then the vector corresponding to each segment is obtained based on the encoding ID, and finally the vectors corresponding to each segment are combined to obtain the vector sequence corresponding to the URL.

[0054] Convolutional networks consider sequence information at a specific window (kernel) granularity. They perform a convolution operation on the vector sequence corresponding to each URL, obtaining a feature vector for each URL, and thus a set of feature vectors corresponding to the set of URLs. Each URL corresponds to one feature vector; when performing a convolution operation on the vector sequence corresponding to a URL, a position in the sliding window corresponds to a feature value in the feature vector.

[0055] For example, such as Figure 2 As shown, the URL of the item image clicked by the user is http: / / xxx.xxx.com / item / jewellery / ctf / 5673425.html, which can be divided into the following four segments: "xxx.xxx.com", "item", "jewellery", and "ctf". The encoded IDs and vectors of these four segments are as follows: Figure 2 As shown in the image.

[0056] When a convolutional neural network performs a convolution operation, assuming the convolution kernel consists of two vectors, a virtual window of size two vectors slides across the vector sequence corresponding to the URL. When the virtual window slides to enclose vectors 1 and 2, the convolution kernel is multiplied by vectors 1 and 2 to obtain feature value 1. Then, the virtual window slides to enclose vectors 2 and 3, and the convolution kernel is multiplied by vectors 2 and 3 to obtain feature value 2. Then, the virtual window slides to enclose vectors 3 and 4, and the convolution kernel is multiplied by vectors 3 and 4 to obtain feature value 3. Therefore, the feature vector corresponding to the URL "http: / / xxx.xxx.com / item / jewellery / ctf / 5673425.html" is the feature vector composed of feature values ​​1, 2, and 3. In this URL, both "jewellery" and "ctf" contain very obvious user preference information. The meaning of "ctf" and "jewellery" appearing together is even more obvious. Therefore, the feature value 3 obtained by performing the inner product operation of vector 3 and vector 4 with the convolution kernel is significantly different from other feature values.

[0057] The input to a feedforward neural network is a set of resources that users pay attention to over a period of time, along with statistical data on the corresponding attention behaviors. For example, this could include the number of clicks and transactions a user makes on financial products or consumer goods within a month, or metrics calculated from log statistics.

[0058] Fully connected layers can be Figure 1 The diagram shows two or more layers, but it can also be a single layer. The outputs of the fully connected layer BERT network, the convolutional neural network, and the feedforward neural network are fused. The activation function of each neuron in the fully connected layer can be the sigmoid function, and the output of the fully connected layer is the probability value corresponding to multiple preset labels.

[0059] Based on the output probability values, a probability threshold can be determined, that is, whether each probability value is greater than or equal to a preset probability threshold. If the determination result is yes, then the preset label corresponding to the probability value that is greater than or equal to the preset probability threshold is determined as the label determined by the user.

[0060] For example, if the preset number of labels is 200, the fully connected layer outputs a 200-dimensional vector, with each dimension corresponding to a label. The sigmoid function will transform each number in the 200-dimensional vector into a probability value between [0,1]. If a probability threshold is set, and the probability value at each position is greater than or equal to this threshold, the corresponding preset label will be assigned to the user.

[0061] Based on the above, this specification provides a label determination method that can be used by resource platform operators or data analysts.

[0062] This resource platform can be a platform that simply displays, searches, and recommends resources, or it can be a platform for resource transactions. The platform can be either an online platform or a physical platform.

[0063] In some embodiments, the platform can also be a combined platform that operates both a web platform and a physical platform, with resources corresponding to those in the web platform and the physical platform. This combined platform can then determine user tags based on data from both the web platform and the physical platform.

[0064] Online platforms can be search engines like Baidu and Google, or shopping platforms like Pinduoduo, Tmall, and JD.com. These platforms can take any of the following forms: websites, mobile applications, or VR (virtual reality) technology.

[0065] Physical platforms, such as the operators of physical stores, can obtain information about users' consumption behaviors in physical stores, such as viewing items, trying (e.g., sampling, trying on clothes), adding items to shopping carts, and checking out. This information can be obtained through manual recording or by recording and analyzing surveillance videos.

[0066] Operators of online or physical platforms can obtain user consumption behavior data based on their platforms and assign tags to users on those platforms.

[0067] Data analysts can obtain data from online platform operators and / or physical store managers, and use this data to determine user tags. (This is in contrast to the previous sentence.) Figure 3 As shown, it includes the following steps:

[0068] S10: Obtain the identifier sequence of resources that the target user is interested in within a predetermined time period.

[0069] In the context of "resources that target users are interested in", the target user's attention behavior towards resources can be at least one of the following for online platforms: clicking, zooming in or viewing in full screen, browsing details, tipping, adding to cart, favorited, reposting, or making a purchase; for physical platforms, it can be at least one of the following: consulting with sales guides, stopping to look, touching and feeling, tasting, trying on, adding to cart, or making a payment.

[0070] In some embodiments, such as Figure 4 As shown, step S10 may include the following steps:

[0071] S11: Obtain the identifiers of each resource that the target user is interested in on the resource platform within a predetermined time period; wherein, the resources on the resource platform are organized by category, and the name or category of the resource is used as the identifier of the resource.

[0072] On online platforms, such as JD.com and Amazon, resources are typically organized by category. Similarly, on physical platforms, such as supermarkets and hypermarkets, resources (e.g., physical goods) are usually organized by category. In this categorized organization, broad categories are further subdivided into smaller categories, and smaller categories are further subdivided into even smaller categories.

[0073] The “resource identifier” mentioned in this specification can be an identifier for these categories. Specifically, it can be an identifier for the smallest category, that is, to locate the content browsed by the user through the identifier for the smallest category; or it can be an identifier for the resource itself. The identifier for the resource itself can also be regarded as an identifier for the category, except that the number of resources in that category is 1.

[0074] The identifiers of resources a user follows on a web platform can be obtained by the electronic device running the web platform, such as by components within the web platform. Similarly, the identifiers of resources a user follows on a physical platform can be manually recorded or obtained by capturing user activity in a physical store using cameras, and then analyzing the captured images to determine the identifiers of the resources the user follows.

[0075] S12: Arrange the identifiers of each resource in the order of their importance to obtain the resource identifier sequence.

[0076] In some embodiments, such as Figure 5 As shown, step S10 may include the following steps:

[0077] S13: Obtain the identifiers of the resources searched by the target user on the resource platform within a predetermined time period, as well as the identifiers of the various resources that the target user follows after the search; wherein, the resources on the resource platform are organized by category, and the name or category of the resource is used as the identifier of the resource.

[0078] On online platforms, such as JD.com and Amazon, resources are typically organized by category. Similarly, on physical platforms, such as supermarkets and hypermarkets, resources (e.g., physical goods) are usually organized by category. In this categorized organization, broad categories are further subdivided into smaller categories, and smaller categories are further subdivided into even smaller categories.

[0079] The term "resource identifier" as used in this specification can refer to identifiers of these categories or to identifiers of the resource itself.

[0080] When browsing a web platform, users may enter search terms in the search bar. These search terms are associated with resource identifiers. Therefore, when obtaining the identifiers of resources that users are interested in on the web platform, these search terms themselves can also serve as resource identifiers, or the resource identifiers associated with the search terms entered by the user can be obtained directly.

[0081] While browsing physical platforms, users may consult sales guides about the location or related information of certain items. The names of these items can be used as search terms. Similarly, when obtaining identifiers of resources that users are interested in on physical platforms, these search terms themselves can also serve as resource identifiers, or resource identifiers that can be directly obtained from these search terms.

[0082] The identifiers of resources a user follows on a web platform can be obtained by the electronic device running the web platform, such as by components within the web platform. Similarly, the identifiers of resources a user follows on a physical platform can be manually recorded or obtained by capturing user activity in a physical store using cameras, and then analyzing the captured images to determine the identifiers of the resources the user follows.

[0083] S14: Sort the identifiers of the searched resources and the identifiers of the browsed resources in chronological order to obtain a resource identifier sequence.

[0084] For example, if a user searches for "wealth management" and then clicks on "premium wealth management," enters the premium wealth management channel, and then clicks on "XX stable wealth management product" and "XX open-ended private banking wealth management product," the corresponding action sequence would be: "wealth management - premium wealth management - XX stable wealth management product - XX open-ended private banking wealth management product."

[0085] S20: Determine the association between adjacent resource identifiers in the identifier sequence of resources, and determine the first vector matrix used to represent the identifier sequence based on the association between adjacent resource identifiers.

[0086] In the sequence of identifiers of resources that the target user is interested in, if there is a relationship between two adjacent resource identifiers, it can be indicated that the user's operation was not accidental. If multiple consecutive resource identifiers are related, it can be indicated that the user is more interested in the resources associated with the consecutively related resource identifiers.

[0087] For example, if the resource identifier sequence is: jewelry - Chow Tai Fook necklace - Chow Tai Fook bracelet - Chow Tai Fook earrings - Chow Tai Fook baby bracelet, then it can be inferred that the user is interested in: jewelry from the Chow Tai Fook brand; if the resource identifier sequence is: jewelry - Cartier necklace - Tiffany necklace - Bulgari diamond ring, then it can be inferred that the user is interested in: luxury jewelry; if the resource identifier sequence is: basketball - jump rope - equipment - treadmill - elliptical machine - rowing machine, then it can be inferred that the user is interested in: sports.

[0088] In some embodiments, the identifier sequence can be input into the BERT network. During the processing of the identifier sequence, the BERT network considers the association between adjacent resource identifiers and determines a first vector matrix to represent the content that the user is interested in based on the association between adjacent resource identifiers.

[0089] In some embodiments, the identifier sequence can also be input into other network models, and the network model outputs a first vector matrix to represent the content that the user is interested in.

[0090] S30: Obtain a set of URLs formed by the URLs of page elements associated with each resource attention behavior of the target user within a predetermined time period, wherein the URL of the page element associated with each resource attention behavior contains at least one resource identifier.

[0091] Step S30 can be applied to scenarios on online platforms. When users browse resources on an online platform, multiple web page elements will be displayed on the page, such as images, videos, music, animations, links, plugins, etc.

[0092] Elements displayed on a webpage typically have a URL (uniform resource locator), representing the network address of the network resource (note that this is not the "resource" in this context) corresponding to the element. These network resources usually reside on a server on the network, where they are typically organized by target, and the directory names are usually related to the category identifiers of the network resources within that directory; these category identifiers are also the resource identifiers. The URL of the network resource corresponding to an element usually includes at least one directory name, meaning the URL contains at least one resource identifier.

[0093] S40: Determine the second vector matrix corresponding to the resources that the target user is interested in within a predetermined time period based on the URL set.

[0094] In some embodiments, resource identifiers can be directly extracted from each URL, and then the resource identifiers can be converted into vectors to obtain vectors corresponding to multiple resource identifiers. These vectors can then be used to form a second vector matrix.

[0095] In some embodiments, S40 may include the following steps:

[0096] S41: Perform the following operations on each URL in the URL set: Divide the current URL into multiple segments according to the preset identifier of the current URL; determine the vector corresponding to each segment, and determine the vector matrix corresponding to the current URL according to the vector corresponding to each segment; perform a convolution operation on the vector matrix corresponding to the current URL to obtain the feature value vector corresponding to the current URL.

[0097] In some embodiments, the vector corresponding to each segment can be determined as follows: the current segment is encoded, and the vector corresponding to the current segment is determined based on the encoding of the current segment. The correspondence between encoding and vector can be pre-set, or the vector corresponding to the encoding of the current segment can be obtained through a pre-trained network model.

[0098] For example, such as Figure 2 As shown, the URL of the item image clicked by the user is http: / / xxx.xxx.com / item / jewellery / ctf / 5673425.html, which can be divided into the following four segments: "xxx.xxx.com", "item", "jewellery", and "ctf". The encoding and vectors of these four segments are as follows: Figure 2 As shown in the image.

[0099] The convolution operation performed on the vector matrix corresponding to the current URL, such as Figure 6 As shown, it may include:

[0100] S601: Sets a sliding window with the same dimension as the preset convolution kernel to slide on the vector matrix corresponding to the current URL.

[0101] S602: Calculate the feature values ​​corresponding to each position when the sliding window slides to the current position with a preset step size in the following way: Perform an inner product operation between the multiple data enclosed by the sliding window when it slides to the current position on the vector matrix corresponding to the current URL and the data in the convolution kernel to obtain the feature values.

[0102] S603: Determine the eigenvalue vector based on the eigenvalues ​​corresponding to multiple positions of the sliding window during the sliding process.

[0103] The above convolution operations can be performed using a convolutional neural network.

[0104] Before being input into the convolutional neural network, each URL in the URL set can undergo the following preprocessing: the URL is divided into multiple segments based on special identifiers (such as " / ") in the URL, each segment is encoded to obtain an encoding ID, and then the vector corresponding to each segment is obtained based on the encoding ID. Finally, the vectors corresponding to each segment are combined to obtain the vector sequence corresponding to the URL.

[0105] by Figure 2 Taking the vector sequence corresponding to the URL shown as an example, assuming the convolution kernel consists of two vectors, a sliding window of size two vectors slides across the vector sequence corresponding to the URL. When the sliding window encompasses vectors 1 and 2, the convolution kernel is multiplied by vectors 1 and 2 to obtain eigenvalue 1. Then, the sliding window encompasses vectors 2 and 3, and the convolution kernel is multiplied by vectors 2 and 3 to obtain eigenvalue 2. Then, the sliding window encompasses vectors 3 and 4, and the convolution kernel is multiplied by vectors 3 and 4 to obtain eigenvalue 3. Therefore, the eigenvalue vector corresponding to the URL "http: / / xxx.xxx.com / item / jewellery / ctf / 5673425.html" is the eigenvalue vector composed of eigenvalue 1, eigenvalue 2, and eigenvalue 3. In this URL, both "jewellery" and "ctf" contain very obvious user preference information. The meaning of "ctf" and "jewellery" appearing together is even more obvious. Therefore, the feature value 3 obtained by performing the inner product operation of vector 3 and vector 4 with the convolution kernel is significantly different from other feature values.

[0106] S42: Determine the eigenvalue vector matrix corresponding to the URL set based on the eigenvalue vectors corresponding to multiple URLs.

[0107] S43: Use the eigenvalue vector matrix as the second vector matrix.

[0108] The steps S41 to S43 above cleverly extract the resources that the user is interested in and the degree of interest in the resources from the URL by performing a convolution operation on the URL.

[0109] S50: Obtain the set of resources that the target user pays attention to within a predetermined time period, as well as the statistical data of the corresponding attention behavior.

[0110] In some embodiments, the statistical data of attention behavior corresponding to a resource includes at least one of the following: the number of clicks on the resource, the browsing time of the resource, the number of times the resource was viewed again, the number of times the resource was saved, the number of times the resource was reposted, and the number of times the resource was liked.

[0111] In some embodiments, statistical data can be processed using a network model, and the processed data can then be fused with the first vector matrix and the second vector matrix. This network model can be a feedforward neural network or other types of neural networks.

[0112] S60: The first vector matrix, the second vector matrix, and statistical data are fused to obtain the user's tags.

[0113] In some embodiments, a fully connected layer can be used to fuse the first vector matrix, the second vector matrix, and statistical data, that is, the first vector matrix, the second vector matrix, and statistical data are input into the fully connected layer.

[0114] A fully connected layer can be one, two, or more layers. The activation function for a fully connected layer can be the sigmoid function, and the output of the fully connected layer is the probability value corresponding to multiple preset labels.

[0115] In some embodiments, a probability judgment function is set after the fully connected layer to determine whether the probability value is greater than or equal to a preset probability threshold. If the judgment result is yes, then the preset label corresponding to the probability value that is greater than or equal to the preset probability threshold is determined as the label determined by the user.

[0116] In some embodiments, the tags for a target user include: determining primary tags and secondary tags for the target user, wherein a secondary tag is a subordinate concept of a primary tag, that is, a primary tag includes multiple secondary tags. In some instances, the tags determined for the target user may also be tertiary, quaternary, or other multi-level tags.

[0117] Two-level tags, such as "Fashionista - Jewelry", "Film & Animation - Movies", "Investment Expert - Stable", where the tag before the hyphen "-" is the first-level tag and the tag after the hyphen "-" is the second-level tag.

[0118] It should be noted that the label determination method provided in this specification may also include only one or two of steps S10, S20, and S30. For example, the label determination method provided in this specification could be... Figure 7 , Figure 8 , Figure 9 , Figure 10 Any one of them.

[0119] The tag determination method, apparatus, and electronic device provided in this specification use a sequence of identifiers of resources of interest to the target user within a predetermined time period to represent the target user's behavioral sequence during that time period. Then, based on the correlation between adjacent resource identifiers in the identifier sequence, a first vector matrix that can be used to identify the identifier sequence is determined. The first vector matrix can be considered as an overview of the series of behaviors represented by the identifier sequence. The tags of the target user determined based on this overview information are less affected by erroneous operations, thus the determined tags are more accurate.

[0120] The tag determination method, apparatus, and electronic device provided in this specification utilize a set of URLs formed by linking page elements to the resource-related behaviors of a target user within a predetermined time period. Based on URL naming rules, resource information is extracted from each URL in the set, and the target user's tags are determined according to this extracted information. Typically, the same resource identifier X may exist in both category A and category B, and categories A and B may be completely different. Therefore, determining the user's interest solely based on resource identifier X is inaccurate. The aforementioned tag determination method extracts the user's interest from the URLs of page elements. Since the URL naming includes resource category information, this method can more accurately determine the user's interest, leading to more accurate tags.

[0121] The tag determination method, apparatus, and electronic device provided in this specification determine user tags based on the identifier sequence of resources followed by the target user within a predetermined time period, the URL set formed by the URLs of page elements associated with each resource follow-up behavior of the target user within the predetermined time period, the set of resources followed by the target user within the predetermined time period, and the statistical data of the follow-up behavior corresponding to the resources. These three factors are comprehensively considered to determine user tags. The data sources vary greatly, which can make up for the defects of each data source itself, and thus the tags of the target users determined based on these data sources are more accurate.

[0122] This specification provides a label determining device that can be used to achieve... Figure 3 The label determination method is shown. For example... Figure 11 As shown, the device includes a first acquisition unit 10, a first determination unit 20, a second acquisition unit 30, a second determination unit 40, and a fusion unit 50.

[0123] The first acquisition unit 10 is used to acquire a sequence of identifiers of resources followed by the target user within a predetermined time period. The first determination unit 20 is used to determine the association relationship between adjacent resource identifiers in the identifier sequence, and determine a first vector matrix to represent the identifier sequence based on the association relationship between adjacent resource identifiers. The second acquisition unit 30 is used to acquire a set of URLs formed by the URLs of page elements associated with each resource follow-up behavior of the target user within the predetermined time period, wherein the URL of the page element associated with each resource follow-up behavior contains at least one resource identifier. The second determination unit 40 is used to determine a second vector matrix corresponding to the resources followed by the target user within the predetermined time period based on the URL set. The fusion unit 50 is used to fuse the first vector matrix and the second vector matrix to obtain tags determined for the target user.

[0124] In some embodiments, the first acquisition unit 10 includes a first acquisition subunit 11 and a first sorting subunit 12.

[0125] The first acquisition subunit 11 is used to acquire the identifiers of each resource browsed by the target user in the resource platform within a predetermined time period; wherein, the resources in the resource platform are organized according to categories, and the name or category of the resource is used as the identifier of the resource. The first sorting subunit 12 is used to arrange the identifiers of each resource according to the order of browsing, to obtain a sequence of resource identifiers.

[0126] In some embodiments, the first sorting subunit 12 includes a second acquisition subunit 121 and a second sorting subunit 122.

[0127] The second acquisition subunit 121 is used to acquire the identifiers of resources searched by the target user in the resource platform within a predetermined time period, as well as the identifiers of various resources that the target user has followed after the search; wherein, the resources in the resource platform are organized according to categories, and the name or category of the resource is used as the identifier of the resource. The second sorting subunit 122 is used to sort the identifiers of the searched resources and the identifiers of the browsed resources in chronological order to obtain a resource identifier sequence.

[0128] In some embodiments, the second determining unit 40 includes a first determining subunit 41, a second determining subunit 42, and a third determining subunit 43.

[0129] The first determining subunit 41 is used to perform the following operations on each URL in the URL set: divide the current URL into multiple segments according to the preset identifier of the current URL; determine the vector corresponding to each segment; determine the vector matrix corresponding to the current URL based on the vectors corresponding to each segment; and perform a convolution operation on the vector matrix corresponding to the current URL to obtain the feature value vector corresponding to the current URL. The second determining subunit 42 is used to determine the feature value vector matrix corresponding to the URL set based on the feature value vectors corresponding to multiple URLs. The third determining subunit 43 is used to use the feature value vector matrix as the second vector matrix.

[0130] In some embodiments, the first determining subunit 41 performs a convolution operation on the vector matrix corresponding to the current URL in the following manner to obtain the feature vector corresponding to the current URL: a sliding window with the same dimension as the preset convolution kernel is set to slide on the vector matrix corresponding to the current URL; the feature values ​​corresponding to the sliding window at each position with a preset step size are calculated in the following manner: the feature values ​​are obtained by performing an inner product operation between the multiple data enclosed by the sliding window at the current position on the vector matrix corresponding to the current URL and the data in the convolution kernel; the feature vector is determined based on the feature values ​​corresponding to the multiple positions of the sliding window during the sliding process.

[0131] In some embodiments, the fusion unit 50 includes a third acquisition subunit 51 and a fusion subunit 52.

[0132] The third acquisition subunit 51 is used to acquire the set of resources followed by the target user within a predetermined time period and the statistical data of the following behavior corresponding to the resources. The fusion subunit 52 is used to fuse the first vector matrix, the second vector matrix and the statistical data to obtain the user's tags.

[0133] In some embodiments, the statistical data of attention behavior corresponding to a resource includes at least one of the following: the number of clicks on the resource, the browsing time of the resource, the number of times the resource was viewed again, the number of times the resource was saved, the number of times the resource was reposted, and the number of times the resource was liked.

[0134] In some embodiments, the target user's tags include primary tags and secondary tags, wherein the primary tags include multiple secondary tags.

[0135] This specification provides a label determining device that can be used to achieve... Figure 7 The label determination method is shown. For example... Figure 12 As shown, the device includes a first acquisition unit 10, a first determination unit 20, and a third determination unit 70.

[0136] The first acquisition unit 10 is used to acquire the identifier sequence of resources that the target user is interested in within a predetermined time period.

[0137] The first determining unit 20 is used to determine the association relationship between adjacent resource identifiers in the identifier sequence, and to determine the first vector matrix used to represent the identifier sequence based on the association relationship between adjacent resource identifiers.

[0138] The third determining unit 70 is used to determine the user's tag based on the first vector matrix.

[0139] This specification provides a label determining device that can be used to achieve... Figure 7 The label determination method is shown. For example... Figure 13 As shown, the device includes a second acquisition unit 30, a second determination unit 40, and a fourth determination unit 80.

[0140] The second acquisition unit 30 is used to acquire a set of URLs formed by the URLs of page elements associated with each resource attention behavior of the target user within a predetermined time period, wherein the URL of the page element associated with each resource attention behavior contains at least one resource identifier.

[0141] The second determining unit 40 is used to determine the second vector matrix corresponding to the resources that the target user is interested in within a predetermined time period based on the URL set.

[0142] The fourth determining unit 80 is used to determine the target user's tag based on the second vector matrix.

[0143] The description and beneficial effects of the above-mentioned label determining device can be found in the description and beneficial effects of the method section, and will not be repeated here.

[0144] This invention also provides an electronic device, such as... Figure 14 As shown, the electronic device may include a processor 1401 and a memory 1402, wherein the processor 1401 and the memory 1402 may be connected via a bus or other means. Figure 14 Taking the example of a connection between China and Israel via a bus.

[0145] Processor 1401 may be a central processing unit (CPU). Processor 1401 may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations thereof.

[0146] Memory 1402, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the tag determination method in this embodiment of the invention (e.g., Figure 11 The first acquisition unit 10, the first determination unit 20, the second acquisition unit 30, the second determination unit 40, and the fusion unit 50 shown are... Figure 12 The first acquisition unit 10, the first determination unit 20, and the third determination unit 70 shown are... Figure 13 (The second acquisition unit 30, the second determination unit 40, and the fourth determination unit 80 are shown). The processor 1401 executes various functional applications and data classification by running non-transitory software programs, instructions, and modules stored in the memory 1402, thereby implementing the tag determination method in the above method embodiments.

[0147] The memory 1402 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor 1401, etc. Furthermore, the memory 1402 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory 1402 may optionally include memory remotely located relative to the processor 1401, and these remote memories may be connected to the processor 1401 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0148] The one or more modules are stored in the memory 1402, and when executed by the processor 1401, they perform the following: Figure 3 , Figures 7 to 10 The label determination method in the illustrated embodiment.

[0149] For specific details regarding the aforementioned electronic devices, please refer to [link / reference]. Figure 3 , Figures 7 to 10 The relevant descriptions and effects in the corresponding embodiments are for reference only and will not be repeated here.

[0150] This specification provides a computer storage medium storing computer program instructions that, when executed by a processor, are implemented... Figure 3 , Figures 7 to 10 The steps of the method shown.

[0151] This specification provides a computer program product comprising a computer program that, when executed by a processor, implements... Figure 3 , Figures 7 to 10 The steps of the method shown.

[0152] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium can also include combinations of the above types of memory.

[0153] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the methodology). However, with technological advancements, many methodological improvements today can be considered direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that a methodological improvement cannot be implemented using hardware physical modules. For example, a Programmable Logic Device (PLD) (e.g., a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program a digital system themselves to "integrate" it onto a PLD, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must also be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog2. Those skilled in the art should also understand that by simply performing some logic programming on the method flow using one of these hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0154] The various embodiments in this specification are described in a progressive manner. For the same or similar parts between the various embodiments, please refer to each other. The focus of each embodiment is to describe the differences from other embodiments.

[0155] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions.

[0156] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0157] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute certain parts of the methods of various embodiments of this application.

[0158] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc.

[0159] This application can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0160] Although this application has been described through embodiments, those skilled in the art will know that this application has many modifications and variations without departing from the spirit of this application, and it is intended that the appended claims cover such modifications and variations without departing from the spirit of this application.

Claims

1. A label determination method characterized by, include: Obtain the identifier sequence of resources that the target user is interested in within a predetermined time period; The association relationship between adjacent resource identifiers in the identifier sequence is determined using a BERT network, and a first vector matrix for representing the identifier sequence is determined based on the association relationship between adjacent resource identifiers. Obtain a set of URLs formed by the URLs of page elements associated with each resource attention behavior of the target user within the predetermined time period, wherein the URL of the page element associated with each resource attention behavior contains at least one resource identifier. Based on the set of URLs, determine the second vector matrix corresponding to the resources that the target user is interested in within the predetermined time period; The first vector matrix and the second vector matrix are input into a fully connected layer for fusion to obtain the tags determined for the target user; The second vector matrix, which determines the association between the target resources of interest to the target user based on the URL set, includes: Perform the following operations on each URL in the URL set: divide the current URL into multiple segments according to the preset identifier of the current URL; determine the vector corresponding to each segment, and determine the vector matrix corresponding to the current URL according to the vector corresponding to each segment; perform a convolution operation on the vector matrix corresponding to the current URL to obtain the feature value vector corresponding to the current URL. Determine the feature value vector matrix corresponding to the URL set based on the feature value vectors corresponding to multiple URLs; The eigenvalue vector matrix is ​​used as the second vector matrix.

2. The method according to claim 1, characterized in that, Obtain the identifier sequence of resources that the target user is interested in within a predetermined time period, including: Obtain the identifiers of each resource browsed by the target user in the resource platform within a predetermined time period; wherein the resources in the resource platform are organized by category, and the name or category of the resource is used as the identifier of the resource; Arrange the identifiers of each resource in the order they were viewed to obtain the resource identifier sequence.

3. The method according to claim 2, characterized in that, Arrange the identifiers of each resource in the order they were viewed to obtain a sequence of resource identifiers, including: The system acquires the identifiers of resources searched by a target user on a resource platform within a predetermined time period, as well as the identifiers of each resource followed by the target user after the search. The resources on the resource platform are organized by category, and the name or category of the resource is used as the identifier of the resource. The identifiers of the searched resources and the identifiers of the browsed resources are sorted in chronological order to obtain a resource identifier sequence.

4. The method according to claim 1, characterized in that, Perform a convolution operation on the vector matrix corresponding to the current URL to obtain the feature vector corresponding to the current URL, including: Set a sliding window with the same dimension as the preset convolution kernel to slide on the vector matrix corresponding to the current URL; The feature values ​​corresponding to the sliding window when it slides to each position with a preset step size are calculated as follows: the feature values ​​are obtained by performing an inner product operation between the multiple data enclosed by the sliding window when it slides to the current position on the vector matrix corresponding to the current URL and the data in the convolution kernel; The feature value vector is determined based on the feature values ​​corresponding to multiple positions of the sliding window during the sliding process.

5. The method according to claim 1, characterized in that, The first vector matrix and the second vector matrix are fused to obtain the tags determined for the target user, including: Obtain the set of resources that the target user pays attention to within the predetermined time period, as well as the statistical data of the attention behavior corresponding to the resources; The user's tags are obtained by fusing the first vector matrix, the second vector matrix, and the statistical data.

6. The method according to claim 5, characterized in that, The statistical data of attention behavior corresponding to the resource includes at least one of the following: number of clicks on the resource, browsing time of the resource, number of times the resource was viewed again, number of times the resource was saved, number of times the resource was reposted, and number of times the resource was liked.

7. The method according to claim 1, characterized in that, Obtain the tags identified for the target user, including: Identify the primary and secondary tags of the target user, where the primary tags include multiple secondary tags.

8. A label determining device, characterized in that, include: The first acquisition unit is used to acquire the identifier sequence of resources that the target user is interested in within a predetermined time period; The first determining unit is configured to use a BERT network to determine the association relationship between adjacent resource identifiers in the identifier sequence, and to determine a first vector matrix for representing the identifier sequence based on the association relationship between adjacent resource identifiers; The second acquisition unit is used to acquire a set of URLs formed by the URLs of page elements associated with each resource attention behavior of the target user within the predetermined time period, wherein the URL of the page element associated with each resource attention behavior contains at least one resource identifier. The second determining unit is used to determine, based on the URL set, the second vector matrix corresponding to the resources that the target user is interested in within the predetermined time period; The fusion unit is used to input the first vector matrix and the second vector matrix into the fully connected layer for fusion to obtain the tags determined for the target user; The second determining unit includes: The first determining subunit is used to perform the following operations on each URL in the URL set: divide the current URL into multiple segments according to the preset identifier of the current URL; determine the vector corresponding to each segment; determine the vector matrix corresponding to the current URL according to the vector corresponding to each segment; perform a convolution operation on the vector matrix corresponding to the current URL to obtain the feature value vector corresponding to the current URL. The second determining subunit is used to determine the feature value vector matrix corresponding to the URL set based on the feature value vectors corresponding to multiple URLs; The third determining subunit is used to use the eigenvalue vector matrix as the second vector matrix.

9. An electronic device, characterized in that, include: A memory and a processor, the processor and the memory being communicatively connected to each other, the memory storing computer instructions, the processor executing the computer instructions to implement the steps of the method according to any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The computer storage medium stores computer program instructions, which, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

11. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the steps of the method according to any one of claims 1 to 7.

Images