Recommendation method, computing device and readable storage medium

Abstract

The present invention discloses a recommendation method, which relates to the field of computers. The recommendation method of the present invention includes: obtaining the business scenario in which the user is currently located. Then, based on the obtained business scenario, the material vector of the associated material and at least one other vector are obtained from a pre-built vector set, the associated material is a material related to the business scenario, and the other vectors include the user vector of the user and the scene vector of the business scenario. Then, based on the material vector of the associated material and at least one other vector, a first fusion vector representing the characteristics of the user and / or a second fusion vector representing the characteristics of each material to be recommended are obtained. Finally, based on the obtained first fusion vector and / or each second fusion vector, the material that the user is interested in is recommended to the user. The present invention uses fusion vectors for recommendation, and the information represented by the fusion vectors is more comprehensive, thereby improving the accuracy of material recommendations.

Description

Technical Field

[0001] The present invention relates to the field of computers, and in particular to a recommendation method, a computing device, and a readable storage medium. Background Art

[0002] With the rapid development of mobile internet, we have entered an era of information explosion. Therefore, selecting content that meets user needs from this vast amount of information is crucial. Against this backdrop, recommendation systems have emerged to alleviate the problem of information overload. However, current recommendation methods fail to consider the correlation between different types of substances when performing recalls, resulting in low accuracy.

[0003] Therefore, a new recommendation method is urgently needed to solve the above problems. Summary of the Invention

[0004] To this end, the present invention provides a recommended method, a computing device, and a readable storage medium to try to solve or at least alleviate the above problems.

[0005] According to one aspect of the present invention, a recommendation method is provided, which includes: obtaining a business scenario in which a user is currently located; based on the business scenario, obtaining a material vector and at least one other vector of an associated material from a pre-constructed vector set, where the associated material is a material related to the business scenario, and the other vectors include a user vector of the user and a scene vector of the business scenario; based on the material vector and at least one other vector of the associated material, obtaining a first fusion vector representing characteristics of the user and / or a second fusion vector representing characteristics of each material to be recommended; and based on the first fusion vector and / or each second fusion vector, recommending materials of interest to the user to the user.

[0006] Optionally, in the recommendation method according to the present invention, based on the business scenario, the material vector and at least one other vector of the associated material are obtained from a pre-constructed vector set, including: when the business scenario is a first type of scenario, the material vector and scenario vector of the associated material are obtained from the vector set, and the first type of scenario is a business scenario with strong correlation with the material; when the business scenario is a second type of scenario, the material vector and user vector of the associated material are obtained from the vector set, and the second type of scenario is a business scenario with strong correlation between materials; when the business scenario is a third type of scenario, the material vector, user vector and scenario vector of the associated material are obtained from the vector set, and the third type of scenario is a business scenario with strong correlation with both materials and users.

[0007] Optionally, in the recommendation method according to the present invention, based on the material vector of the associated material and at least one other vector, a first fusion vector characterizing the characteristics of the user and / or a second fusion vector characterizing the characteristics of each material to be recommended is obtained, including: when the business scenario is a first type of scenario, the material vector of each material to be recommended is fused with the scenario vector respectively to obtain each second fusion vector, and the material to be recommended is the material in the associated materials other than the target material currently browsed by the user.

[0008] Optionally, in the recommendation method according to the present invention, materials of interest to the user are recommended to the user based on the first fusion vector and / or each second fusion vector, including: calculating the similarity between the material vector of the target material and each second fusion vector; and recommending the materials of interest to the user based on the obtained similarity.

[0009] Optionally, in the recommendation method according to the present invention, based on the material vector of the associated material and at least one other vector, a first fusion vector characterizing the characteristics of the user and / or a second fusion vector characterizing the characteristics of each material to be recommended is obtained, including: when the business scenario is the second type of scenario, the user vector is fused with the material vectors of all reference materials to obtain a first fusion vector, and the reference material is a material in the associated material that has an associated relationship with the user.

[0010] Optionally, in the recommendation method according to the present invention, materials of interest to the user are recommended to the user based on the first fusion vector and / or each second fusion vector, including: calculating the similarity between the first fusion vector and the material vector of each material to be recommended, where the material to be recommended is the material in the associated materials other than the reference material; based on the obtained similarity, the material of interest to the user is recommended to the user.

[0011] Optionally, in the recommendation method according to the present invention, based on the material vector of the associated material and at least one other vector, a first fusion vector characterizing the characteristics of the user and / or a second fusion vector characterizing the characteristics of each material to be recommended is obtained, including: when the business scenario is a third type of scenario, fusing the user vector with the scene vector to obtain a first fusion vector; and fusing the material vector of each associated material with the scene vector respectively to obtain each second fusion vector.

[0012] Optionally, in the recommendation method according to the present invention, materials of interest to the user are recommended to the user based on the first fusion vector and / or each second fusion vector, including: calculating the similarity between the first fusion vector and each second fusion vector; and recommending materials of interest to the user based on the obtained similarity.

[0013] Optionally, in the recommendation method according to the present invention, the step of constructing a vector set includes: constructing a heterogeneous graph based on the user's historical behavior, where the nodes in the heterogeneous graph include users, materials and business scenarios; wandering in the heterogeneous graph according to the determined wandering mode and wandering weight to generate multiple sequences of preset lengths; obtaining training data from the generated sequences, and training the vector generation model based on the training data; obtaining the weight matrix of the hidden layer from the trained vector generation model; and determining the user vector of each user in the heterogeneous graph, the material vector of each material, and the scene vector of each business scenario according to the weight matrix to form a vector set.

[0014] Optionally, in the recommendation method according to the present invention, the vector generation model is a word2vec model.

[0015] Optionally, in the recommendation method according to the present invention, an element-wise approach or a weighted approach is adopted to obtain the fusion vector.

[0016] According to another aspect of the present invention, a computing device is provided, comprising: at least one processor; and a memory storing program instructions, wherein the program instructions are configured to be suitable for execution by the at least one processor, and the program instructions include instructions for executing the recommended method according to the present invention.

[0017] According to another aspect of the present invention, a readable storage medium storing program instructions is provided. When the program instructions are read and executed by a computing device, the computing device executes the recommendation method according to the present invention.

[0018] According to the recommendation method of the present invention, the business scenario in which the user is currently located is first obtained, and then based on the obtained business scenario, the material vector and at least one other vector of the associated material are obtained from the pre-built vector set, and then based on the material vector and at least one other vector of the associated material, a first fusion vector representing the characteristics of the user and / or a second fusion vector representing the characteristics of each material to be recommended is obtained. Finally, based on the first fusion vector and / or each second fusion vector, the material that the user is interested in is recommended to the user. It can be seen that the present invention uses a fusion vector for recommendation, and the fusion vector can better capture the characteristics of the user and the characteristics of the material to be recommended, and the information it represents is more comprehensive, thereby improving the accuracy of the material recommendation. In addition, the present invention uses a corresponding fusion vector based on the current business scenario type, so that it can accurately represent the characteristics of the user and the characteristics of the material to be recommended, and can also reduce a certain amount of calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] To achieve the above and related purposes, certain illustrative aspects are described herein in conjunction with the following description and accompanying drawings, which indicate various ways in which the principles disclosed herein may be practiced, and all aspects and their equivalents are intended to fall within the scope of the claimed subject matter. The above and other objects, features, and advantages of the present disclosure will become more apparent by reading the following detailed description in conjunction with the accompanying drawings. Throughout this disclosure, the same reference numerals generally refer to the same parts or elements.

[0020] Figure 1 A schematic diagram showing a recommended method according to an embodiment of the present invention is shown;

[0021] Figure 2 shows a structural block diagram of a computing device 200 according to one embodiment of the present invention;

[0022] Figure 3 shows a flow chart of a recommendation method 300 according to one embodiment of the present invention;

[0023] Figure 4 A schematic diagram of a constructed heterogeneous graph according to an embodiment of the present invention is shown;

[0024] Figure 5 A schematic diagram of a generated walk sequence according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0025] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0026] Current recommendation methods do not consider the correlation between different types of substances when recalling, and the accuracy is low. Based on this, the present invention provides a recommendation method that uses a fusion vector for recall. Specifically, first, a heterogeneous graph is constructed based on the user's historical behavior. Then, based on the constructed heterogeneous graph, a sequence of a certain length is generated, and model training is performed based on the generated sequence. Finally, the trained model is used to obtain the characterization vectors of multiple types of substances and store them in a container. Then, when making recommendations later, the required material vectors are first obtained from the container based on the current business scenario, and then fused. Finally, the materials are recalled based on the fusion vector and recommended to the user. Among them, the recommendation method of the present invention can be executed in a computing device.

[0027] Figure 2A block diagram of the physical components (i.e., hardware) of a computing device 200 is shown. In a basic configuration, the computing device 200 includes at least one processing unit 202 and system memory 204. According to one aspect, depending on the configuration and type of the computing device, the processing unit 202 can be implemented as a processor. The system memory 204 includes, but is not limited to, volatile storage (e.g., random access memory), non-volatile storage (e.g., read-only memory), flash memory, or any combination of such memories. According to one aspect, the system memory 204 includes an operating system 205 and program modules 206. The program modules 206 include a recommendation module 220. The recommendation module 220 is configured to perform the recommendation method 300 of the present invention.

[0028] According to one aspect, operating system 205 is suitable for controlling the operation of computing device 200, for example. Furthermore, examples may be practiced in conjunction with graphics libraries, other operating systems, or any other application programs, and are not limited to any particular application or system. Figure 2 This basic configuration is illustrated in FIG. 2 by those components within dashed line 208. According to one aspect, computing device 200 has additional features or functionality. For example, according to one aspect, computing device 200 includes additional data storage devices (removable and / or non-removable), such as magnetic disks, optical disks, or tapes. Such additional storage Figure 2 2 is illustrated by removable storage 209 and non-removable storage 210.

[0029] As stated above, according to one aspect, program modules are stored in the system memory 204. According to one aspect, the program modules may include one or more application programs, and the present invention is not limited to the type of application program. For example, the application program may include an email and contact application program, a word processing application program, a spreadsheet application program, a database application program, a slide show application program, a drawing or computer-aided application program, a web browser application program, etc.

[0030] According to one aspect, examples may be practiced on a circuit comprising discrete electronic components, a packaged or integrated electronic chip containing logic gates, a circuit utilizing a microprocessor, or a single chip containing electronic components or a microprocessor. Figure 2In the embodiment of the present invention, each or many components shown in the embodiment of the present invention can be integrated into a system on a chip (SOC) on a single integrated circuit to practice the example. According to one aspect, such a SOC device may include one or more processing units, a graphics unit, a communication unit, a system virtualization unit, and various application functions, all of which are integrated (or "burned") onto a chip substrate as a single integrated circuit. When operated via the SOC, the functions described in this article can be operated via a dedicated logic integrated with other components of the computing device 200 on a single integrated circuit (chip). Other technologies capable of performing logical operations (such as AND, OR, and NOT) can also be used to practice embodiments of the present invention, and the other technologies include but are not limited to mechanical, optical, fluid, and quantum technologies. In addition, embodiments of the present invention can be practiced in a general-purpose computer or in any other circuit or system.

[0031] According to one aspect, the computing device 200 may also have one or more input devices 212, such as a keyboard, a mouse, a pen, a voice input device, a touch input device, etc. Output devices 214 may also be included, such as a display, speakers, a printer, etc. The aforementioned devices are examples, and other devices may also be used. The computing device 200 may include one or more communication connections 216 that allow communication with other computing devices 218. Examples of suitable communication connections 216 include, but are not limited to: RF transmitter, receiver, and / or transceiver circuitry; Universal Serial Bus (USB), parallel, and / or serial ports.

[0032] The term computer-readable medium as used herein includes computer storage media. Computer storage media can include volatile and non-volatile, removable and non-removable media implemented with any method or technology for storing information (e.g., computer-readable instructions, data structures, or program modules). System memory 204, removable storage 209, and non-removable storage 210 are all examples of computer storage media (i.e., memory storage). Computer storage media can include random access memory (RAM), read-only memory (ROM), electrically erasable read-only memory (EEPROM), flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical storage, cassette tape, magnetic tape, disk storage or other magnetic storage device, or any other product that can be used to store information and can be accessed by computing device 200. According to one aspect, any such computer storage medium can be a part of computing device 200. Computer storage media does not include carrier waves or other propagated data signals.

[0033] According to one aspect, communication media is implemented by computer-readable instructions, data structures, program modules, or other data in a modulated data signal (e.g., a carrier wave or other transport mechanism), and includes any information delivery media. According to one aspect, the term "modulated data signal" describes a signal that has one or more characteristics set or changed in such a manner as to encode information in the signal. By way of example, and not limitation, communication media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, radio frequency (RF), infrared, and other wireless media.

[0034] Figure 3 A method 300 is shown according to an embodiment of the present invention, which is suitable for use in a computing device (eg Figure 2 Executed in the computing device 200 shown in FIG. Figure 3 As shown, the method 300 begins at 310 .

[0035] At 310 , the user's current business scenario is obtained. Business scenarios can be categorized into three types: business scenarios with strong material relevance, business scenarios with strong material-to-material relevance, and business scenarios with strong relevance to both materials and users. For ease of description, business scenarios with strong material relevance will be referred to as first-type scenarios, business scenarios with strong material-to-material relevance will be referred to as second-type scenarios, and business scenarios with strong relevance to both materials and users will be referred to as third-type scenarios.

[0036] According to one embodiment of the present invention, the first type of scenario may include a content-related recommendation scenario (i.e., recommending content related to the content to the user, such as the content-related recommendation scenario of the Zhijia app. When the user views the BMW configuration content, the user may be recommended with content such as the BMW driving experience and BMW price quotes), a car series-related recommendation scenario (i.e., recommending car series related to the car series to the user, such as the car series-related recommendation scenario of the Zhijia app. When the user views the introduction of the BMW 320 car series, the user may be recommended with car series cards such as the C200 and Audi A4, guiding the user to give clues to the 4S store), a second type of scenario may include a follow page (comprehensively considering the user and the author followed by the user, recommending authors that may be of interest to the user), a live broadcast page (comprehensively considering the user and the anchor followed by the user, recommending anchors that may be of interest to the user), a third type of scenario may include a recommendation home page (such as the Zhijia app recommendation home page, recommending personalized car information, configuration, evaluation, etc. to the user), a search result page (such as the Zhijia app search result page, comprehensively considering the search terms and users to provide users with relevant and personalized car-related content). Of course, this is only an example and is not limited to this. In addition, for different types of business scenarios, in some embodiments, they can be marked with different identifiers. In this way, after obtaining the business scenario in which the user is currently located, the type of the current business scenario can be determined by reading its identifier.

[0037] Then, proceeding to 320 , based on the acquired business scenario, the material vector of the associated material and at least one other vector are acquired from the pre-built vector set. Here, the associated material, other vectors, and vector set are first described.

[0038] Each business scenario has its associated (and corresponding) materials, where the associated materials refer to materials related to the current business scenario. Other vectors, according to one embodiment of the present invention, may include a user vector of the user and a scenario vector of the business scenario.

[0039] As for the vector set, in some embodiments, it can be constructed in the following way. First, a heterogeneous graph is constructed based on the user's historical behavior. Specifically, according to the user's historical behavior, the connection relationship between the nodes is determined, and then a heterogeneous relationship connection undirected graph is constructed by the nodes and the connection relationship. Among them, the user's historical behavior may include the user's click, browse, follow and other behaviors, and the nodes in the heterogeneous graph may include users, materials and business scenarios, which are not specifically limited in the present invention. In addition, in order to better understand the construction of the heterogeneous graph, an example is given below, such as Figure 4 , which consists of three types of nodes: users, business scenarios, and materials, and three connection relationships: entry (users enter scenarios), distribution (scenario distributes materials), and click (users click on materials), forming a heterogeneous graph.

[0040] Subsequently, using the determined walk pattern and walk weight, a walk is performed within the constructed heterogeneous graph to generate multiple sequences of preset lengths. The walk pattern and walk weight can be determined by technical personnel based on business objectives. For example, if the business objective is to increase the time users spend on the recommended homepage, the walk pattern can be a cyclic weighted walk from user to business scenario to material, where users move from the business scenario to the material, and then from the material to the user, repeating the cycle. Specifically, the walk weight from user to business scenario can be determined by the importance of the scenario. For example, if the recommended homepage is significantly more important than the travel page, the walk weight from the business scenario to the material can be determined by the number of times the material is distributed by the scenario, and the walk weight from the material to the user can be determined by the time users spend consuming the material. Based on the walk weights, multiple cyclic walks can be performed within the heterogeneous graph using the determined walk pattern to generate multiple sequences of preset lengths. Each walk can prioritize objects with significant walk weights, and the diversity of the generated sequences can be increased by increasing the number of cyclic walks. The present invention does not limit the length of the sequence, i.e., the preset length. For example, it can be 5, 10, etc. In a specific embodiment, those skilled in the art can set it according to actual needs. In addition, in order to better understand this step, the present invention continues to use the above example as an example to provide a schematic diagram of two walk sequences with a length of 7 generated after two walks with users A1 and A3 as the starting points, as shown in FIG. Figure 5 .

[0041] Next, training data is obtained from the generated sequence, and the vector generation model is trained based on the obtained training data. In some embodiments, the training data can be obtained by intercepting a subsequence of a preset window length from the generated sequence. Specifically, for any sequence, each consecutive window length of data is used as a training data. For example, if the sequence length is 10, the content is 1, 2, 3...10, and the window length is 5, then the subsequence composed of every 5 consecutive contents can be used as a training data, such as 1, 2, 3, 4, 5 can be used as a training data, and 2, 3, 4, 5, 6 can also be used as a training data.

[0042] For each window length of training data, the window center data and the data on both sides of the window center can be combined into a data pair, and each data pair can be used as a training sample to train the vector generation model. For example, using the training data 1, 2, 3, 4, and 5 shown above as an example, the window center data 3 and the data 1, 2, 4, and 5 on both sides of the window center can be combined into a data pair, and this data pair can be used as a training sample to train the vector generation model.

[0043] According to one embodiment of the present invention, a vector generation model can be trained using the window center data as input and the data on both sides of the window center as output. Specifically, the window center data is input into the vector generation model to obtain the predicted data on both sides of the window center. Then, based on the loss value between the predicted data on both sides of the window center and the actual data on both sides of the window center, the vector generation model is updated until the loss value meets a predetermined condition. The training is then completed, and a trained vector generation model is obtained. In some embodiments, the loss value between the predicted data on both sides of the window center and the actual data on both sides of the window center can be the sum of the loss values ​​between each piece of data on both sides of the predicted window center and each piece of data on both sides of the actual window center.

[0044] That is to say, in this embodiment, the vector generation model is trained by inferring the data on both sides of the window by the data at the center of the window, and the loss function is the sum of the loss functions of the data on both sides of the window. Furthermore, considering that each prediction must be calculated based on the entire data set, which will bring a lot of time overhead, a downsampling strategy can be used to optimize the training speed during the training process. Each time, part of the data is selected as a negative sample to update the weight. The part of the data is determined by the frequency of occurrence in the sample. Data with a high frequency of occurrence is more likely to be selected as a negative sample. Conversely, low-frequency data is less likely to be selected. In addition, in some embodiments, the vector generation model can adopt the word2vec model, and the training method can adopt the skipgram method. Of course, this is just an example, and the present invention is not limited to this.

[0045] At this point, the training of the vector generation model is complete. Subsequently, the hidden layer weight matrix is ​​obtained from the trained vector generation model. Specifically, the weight matrix from the input layer to the hidden layer, commonly referred to as the word vector matrix, is obtained. In some embodiments, the number of rows in this weight matrix can be the sum of the number of users, the number of materials, and the number of business scenarios, and the number of columns can be the latitude of the vector.

[0046] Finally, based on the obtained weight matrix, the user vector of each user in the heterogeneous graph, the material vector of each material, and the scene vector of each business scenario are determined, and a vector set is formed. Specifically, each row in the weight matrix is ​​used as a vector of the object (i.e., user, material, or business scenario) to which it corresponds, and then the vector is used as an element of the vector set. That is, the vector set is composed of the user vector of each user, the material vector of each material, and the scene vector of each business scenario. Furthermore, in some embodiments, after obtaining the user vector of each user, the material vector of each material, and the scene vector of each business scenario, the vector set can be constructed by storing them in a vector container.

[0047] After clearly understanding the associated material, other vectors, and vector sets, the following describes how to obtain the associated material's material vector and at least one other vector from a pre-built vector set based on the acquired business scenario. Specifically, when the acquired business scenario is a first-type scenario, the associated material's material vector and the current business scenario's scenario vector are obtained from the vector set. When the business scenario is a second-type scenario, the associated material's material vector and user vector are obtained from the vector set. When the business scenario is a third-type scenario, the associated material's material vector, user vector, and the current business scenario's scenario vector are obtained from the vector set.

[0048] Then, the process proceeds to 330 , where a first fusion vector representing the characteristics of the user and / or a second fusion vector representing the characteristics of each to-be-recommended material is obtained based on the obtained material vector of the associated material and at least one other vector.

[0049] When the business scenario is a first-type scenario, the item vectors of each item to be recommended are fused with the scenario vector to obtain second fused vectors. The items to be recommended are those in the associated items other than the target item currently being viewed by the user. That is, when the current business scenario is highly material-related, all items in the associated items related to the current business scenario other than the item currently being viewed by the user are considered as items to be recommended. The item vectors of each item to be recommended are fused with the scenario vector to form a second fused vector representing the characteristics of each item to be recommended. For example, when the business scenario is a car series recommendation scenario (in this case, the item is a car series, which refers to a model of a certain brand, such as BMW, Mercedes-Benz, Audi, etc.), all car series related to the car series recommendation scenario other than the car series currently being viewed by the user are considered as car series to be recommended. The vectors of each car series to be recommended (i.e., car series vectors) are then fused with the vectors of the car series recommendation scenario (i.e., scene vectors) to form a second fused vector representing the characteristics of each car series to be recommended. Furthermore, in some embodiments, an element-wise or weighted approach can be used to fuse the material vector and scene vector of the recommended material. That is, the second fused vector can be obtained using various element-wise pooling methods, or by assigning different weights to the material vector and scene vector. Of course, other methods can also be used to obtain the second fused vector, and this invention is not limited to these methods.

[0050] When the business scenario is the second type, the user vector is fused with the material vectors of all reference materials to obtain a first fused vector, where the reference materials are the materials in the associated materials that are already associated with the user. That is, when the current business scenario is one with strong material-to-material correlation, the materials in the associated materials that are currently associated with the user are used as reference materials, and the remaining materials are used as materials to be recommended. The user vector is then fused with the material vectors of all reference materials to form the first fused vector that characterizes the user's characteristics. This association is related to the business scenario. For example, when the business scenario is a focus scenario, the association can be focus, and the reference materials are the materials in the associated materials that the user has already followed. The following example uses a focus scenario (e.g., a page to be followed) as an example. Specifically, all materials related to the focus scenario (e.g., an author) except for those that the user has already followed are used as materials to be recommended, and the materials that the user has already followed are used as reference materials. The user vector is then fused with the reference materials to form the first fused vector that characterizes the user's characteristics. The acquisition of the first fusion vector is similar to the acquisition of the second fusion vector, and an element-wise approach or a weighted approach may be used.

[0051] When the business scenario is a third type of scenario, the user vector is fused with the scenario vector to obtain a first fused vector, and the material vector of each associated material is fused with the scenario vector to obtain each second fused vector. That is, when the current business scenario is a business scenario with strong correlation with both materials and users, the user vector is fused with the scenario vector to form a first fused vector that characterizes the user's characteristics, and the associated materials related to the current business scenario are all used as materials to be recommended, and each material to be recommended is fused with the scenario vector to obtain a second fused vector that characterizes each material to be recommended. The following example uses the case where a user enters a recommendation homepage (the recommendation homepage belongs to the third type of scenario). Specifically, the user vector is fused with the scenario vector of the recommendation homepage to form a first fused vector that characterizes the user's characteristics, and each associated material related to the recommendation homepage is used as a material to be recommended, and each material to be recommended is fused with the scenario vector of the recommendation homepage to form a second fused vector that characterizes the characteristics of each material to be recommended. As described above, the acquisition of the first fused vector and the second fused vector can be performed using an element-wise approach or a weighted approach, which will not be elaborated here. In addition, in some embodiments, when the business scenario is a third type scenario, when obtaining a first fusion vector representing the user's characteristics and a second fusion vector representing the characteristics of each material to be recommended, some other vectors may also be fused. For example, when the business scenario is a search results page, the user vector may be fused with the scenario vector and the search term vector to obtain a first fusion vector, and the material vectors of each associated material may be fused with the scenario vector and the search term vector to obtain each second fusion vector. Two further points are explained here. First, the other vectors in this case also include the search term vector; second, for scenarios in the third type scenario where other vectors such as search terms need to be fused, in some embodiments, they can be distinguished by marking a special identifier.

[0052] As can be seen, in this embodiment, when the business scenario is a first type scenario, a second fusion vector representing the characteristics of each item to be recommended is obtained; when the business scenario is a second type scenario, a first fusion vector representing the characteristics of the user is obtained; and when the business scenario is a third type scenario, both the first fusion vector representing the characteristics of the user and the second fusion vector representing the characteristics of each item to be recommended are obtained. In other words, for the first type scenario, this embodiment uses a fusion vector to represent the characteristics of the item to be recommended; for the second type scenario, this embodiment uses a fusion vector to represent the characteristics of the user; and for the third type scenario, this embodiment uses a fusion vector to represent both the characteristics of the user and the characteristics of each item to be recommended.

[0053] At this point, the fusion vector (i.e., the first fusion vector or the second fusion vector) is obtained. Then, the process proceeds to 340 , where materials of interest to the user are recommended to the user based on the obtained first fusion vector and / or each second fusion vector. The details are as follows.

[0054] When the business scenario is the first type of scenario, the similarity between the material vector of the target material currently browsed by the user and the second fusion vector representing the characteristics of each material to be recommended is first calculated. Among them, the similarity can be calculated using the Euclidean distance or cosine method. Then, based on the obtained similarity, the material that the user is interested in is recommended to the user. Specifically, the similarities can be sorted in descending order, and the recommended materials corresponding to the predetermined number of similarities at the front are recommended to the user as the materials that the user is interested in. Of course, this is just an example, and the present invention is not limited to this. In addition, the present invention is not limited to the predetermined number. In a specific embodiment, those skilled in the art can set it according to actual needs.

[0055] When the business scenario falls under the second type, the similarity between the first fused vector representing the user's features and the item vectors of each item to be recommended is first calculated. Then, based on the obtained similarity, items of interest are recommended to the user. The method for recommending items of interest to the user based on the obtained similarity can be found in the first type of scenario described above and will not be further described here.

[0056] When the business scenario falls under the third type, the similarity between the first fused vector representing the user's characteristics and the second fused vector representing the characteristics of each item to be recommended is calculated. Then, based on the obtained similarity, items of interest to the user are recommended to the user. The method for recommending items of interest to the user based on the obtained similarity can be found in the first type of scenario described above and will not be further described here.

[0057] As can be seen from this, in the present invention, the first type of scenario uses a single vector-fused vector recall method, the second type of scenario uses a fused vector-single vector recall method, and the third type of scenario uses a fused vector-fused vector recall method. In other words, the present invention uses a fused vector for recommendation. The fused vector can better capture the characteristics of the user and the characteristics of the material to be recommended, and the information it represents is more comprehensive, thereby improving the accuracy of material recommendations. In addition, the present invention uses the corresponding fused vector based on the current business scenario type, which can not only accurately characterize the characteristics of the user and the characteristics of the material to be recommended, but also reduce a certain amount of calculation.

[0058] Furthermore, in this invention, each vector is derived from a heterogeneous graph, thus representing a more comprehensive picture. Furthermore, the invention can adjust the walk method and walk weight based on business objectives, thus generating more targeted training data and further enhancing the comprehensiveness and accuracy of the information represented by each vector.

[0059] A9. A method as described in any one of A1-A8, wherein the step of constructing the vector set includes: constructing a heterogeneous graph based on the user's historical behavior, wherein the nodes in the heterogeneous graph include users, materials, and business scenarios; wandering in the heterogeneous graph according to the determined wandering mode and wandering weight to generate multiple sequences of preset lengths; obtaining training data from the generated sequences, and training the vector generation model based on the training data; obtaining the weight matrix of the hidden layer from the trained vector generation model; and determining the user vector of each user in the heterogeneous graph, the material vector of each material, and the scene vector of each business scenario according to the weight matrix to form the vector set. A10. A method as described in A9, wherein the vector generation model is a word2vec model. A11. A method as described in any one of A1-A10, wherein the fusion vector is obtained using an element-wise method or a weighted method.

[0060] The various techniques described herein may be implemented in conjunction with hardware or software, or a combination thereof. Thus, the methods and apparatus of the present invention, or certain aspects or portions of the methods and apparatus of the present invention, may be implemented in the form of program codes (i.e., instructions) embedded in a tangible medium, such as a removable hard disk, a USB flash drive, a floppy disk, a CD-ROM, or any other machine-readable storage medium, wherein when the program is loaded into a machine such as a computer and executed by the machine, the machine becomes an apparatus for practicing the present invention.

[0061] When program code is executed on a programmable computer, the computing device generally includes a processor, a storage medium readable by the processor (including volatile and non-volatile memory and / or storage elements), at least one input device, and at least one output device. The memory is configured to store the program code, and the processor is configured to execute the recommended method of the present invention according to instructions in the program code stored in the memory.

[0062] By way of example and not limitation, readable media include readable storage media and communication media. Readable storage media store information such as computer-readable instructions, data structures, program modules, or other data. Communication media typically embody computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and include any information delivery medium. Combinations of any of the above are also included within the scope of readable media.

[0063] In the description provided herein, the algorithms and displays are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems may also be used in conjunction with the examples of the present invention. Based on the above description, it is apparent that the structure required for constructing such systems is well understood. In addition, the present invention is not directed to any specific programming language. It should be understood that various programming languages ​​may be utilized to implement the present invention described herein, and the description of specific languages ​​above is provided for the purpose of disclosing the preferred embodiment of the present invention.

[0064] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures, and techniques are not shown in detail so as not to obscure the understanding of this description.

[0065] It should be understood that in order to streamline the present disclosure and aid understanding of one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, various features of the present invention are sometimes grouped together in a single embodiment, figure, or description thereof. However, this method of disclosure should not be interpreted as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim.

[0066] Those skilled in the art will appreciate that the modules, units, or components of the devices in the examples disclosed herein may be arranged in the device described in the embodiment, or alternatively may be located in one or more devices different from the devices in the examples. The modules in the foregoing examples may be combined into one module or further divided into multiple submodules.

[0067] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and in addition may be divided into multiple submodules or subunits or subcomponents. All features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed herein may be combined in any combination, except that at least some of such features and / or processes or units are mutually exclusive. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.

[0068] Furthermore, those skilled in the art will appreciate that although some embodiments described herein include certain features and not other features included in other embodiments, the combination of features from different embodiments is intended to be within the scope of the invention and to form different embodiments.

[0069] In addition, some of the embodiments are described herein as methods or combinations of method elements that can be implemented by a processor of a computer system or by other devices that perform the functions described. Thus, a processor having the necessary instructions for implementing the method or method element forms a device for implementing the method or method element. Furthermore, the elements described herein of the device embodiments are examples of devices for implementing the functions performed by the elements for the purpose of implementing the invention.

[0070] As used herein, unless otherwise specified, the use of ordinal numbers "first," "second," "third," etc. to describe common objects merely indicates that different instances of similar objects are involved and are not intended to imply that the objects so described must have a given order in time, space, ranking, or in any other manner.

[0071] Although the present invention has been described with respect to a limited number of embodiments, it will be apparent to those skilled in the art, having benefit of the foregoing description, that other embodiments are contemplated within the scope of the invention thus described. Furthermore, it should be noted that the language used in this specification has been selected primarily for readability and didactic purposes, rather than for the purpose of explaining or limiting the subject matter of the present invention. Consequently, many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the appended claims. The disclosure of the present invention is intended to be illustrative rather than restrictive of the scope of the invention, which is defined by the appended claims.

Claims

1. A recommendation method, comprising: Get the user's current business scenario; Based on the business scenario, obtaining a material vector of an associated material and at least one other vector from a pre-built vector set, wherein the associated material is a material related to the business scenario, and the other vectors include a user vector of the user and a scenario vector of the business scenario; Based on the material vector of the associated material and at least one other vector, obtaining a first fusion vector representing characteristics of the user and / or a second fusion vector representing characteristics of each material to be recommended; recommending materials of interest to the user based on the first fusion vector and / or the second fusion vectors; Among them, the steps of constructing the vector set include: constructing a heterogeneous graph based on the user's historical behavior, wherein the nodes in the heterogeneous graph include users, materials and business scenarios; wandering in the heterogeneous graph according to the determined wandering mode and wandering weight to generate multiple sequences of preset lengths; obtaining training data from the generated sequences, and training the vector generation model based on the training data; obtaining the weight matrix of the hidden layer from the trained vector generation model; and determining the user vector of each user in the heterogeneous graph, the material vector of each material, and the scene vector of each business scenario according to the weight matrix to form the vector set.

2. The method according to claim 1, wherein The acquiring of the material vector and at least one other vector of the associated material from a pre-built vector set based on the business scenario includes: When the business scenario is a first type scenario, obtaining a material vector and a scenario vector of an associated material from the vector set, wherein the first type scenario is a business scenario with a strong correlation with a material; When the business scenario is a second type scenario, obtaining a material vector and a user vector of an associated material from the vector set, wherein the second type scenario is a business scenario with strong correlation between materials; When the business scenario is a third type scenario, the material vector, user vector, and scenario vector of the associated material are obtained from the vector set. The third type scenario is a business scenario with strong correlation with both materials and users.

3. The method according to claim 2, wherein: The acquiring, based on the material vector of the associated material and at least one other vector, a first fusion vector representing the characteristics of the user and / or a second fusion vector representing the characteristics of each to-be-recommended material includes: When the business scenario is a first type scenario, the material vectors of the materials to be recommended are respectively fused with the scenario vector to obtain the second fusion vectors, and the materials to be recommended are the materials in the associated materials except the target material currently browsed by the user; Recommending materials of interest to the user based on the first fusion vector and / or the second fusion vectors includes: Calculating the similarity between the material vector of the target material and each of the second fusion vectors; Based on the obtained similarity, materials that the user is interested in are recommended to the user.

4. The method according to claim 2 or 3, wherein: The acquiring, based on the material vector of the associated material and at least one other vector, a first fusion vector representing the characteristics of the user and / or a second fusion vector representing the characteristics of each to-be-recommended material includes: When the business scenario is the second type scenario, the user vector is fused with the material vectors of all reference materials to obtain the first fused vector, where the reference material is a material that has an associated relationship with the user among the associated materials; The recommending materials of interest to the user based on the first fusion vector and / or the second fusion vectors includes: Calculating the similarity between the first fusion vector and the material vectors of each to-be-recommended material, where the to-be-recommended material is the material in the associated materials excluding the reference material; Based on the obtained similarity, materials that the user is interested in are recommended to the user.

5. The method according to claim 2 or 3, wherein: The acquiring, based on the material vector of the associated material and at least one other vector, a first fusion vector representing the characteristics of the user and / or a second fusion vector representing the characteristics of each to-be-recommended material includes: When the business scenario is a third type scenario, fusing the user vector with the scenario vector to obtain the first fused vector; and Fusing the material vectors of the associated materials with the scene vectors to obtain the second fused vectors; The recommending materials of interest to the user based on the first fusion vector and / or the second fusion vectors includes: Calculating the similarity between the first fused vector and each of the second fused vectors; Based on the obtained similarity, materials that the user is interested in are recommended to the user.

6. The method according to any one of claims 1 to 3, wherein The vector generation model is a word2vec model.

7. The method according to any one of claims 1 to 3, wherein The fusion vector is obtained using an element-wise or weighted method.

8. A computing device comprising: at least one processor; as well as A memory storing program instructions, wherein the program instructions are configured to be executed by the at least one processor, and the program instructions include instructions for executing the method according to any one of claims 1 to 7.

9. A readable storage medium storing program instructions, wherein when the program instructions are read and executed by a computing device, the computing device is caused to execute the method according to any one of claims 1 to 7.

Images