Object recommendation method and related device

Abstract

The embodiment of the present application discloses an object recommendation method and related equipment, which are used to generate an object recommendation list based on user historical operation data. The method of the embodiment of the present application includes: obtaining a preset target operation and historical operation data of a target user, wherein each piece of the historical operation data is an operation performed by the target user on an existing object; determining the co-occurrence matrix of the preset target operation and each type of operation in the historical operation data respectively according to the historical operation data; calculating the weight coefficient of each type of operation in the multiple types of operations according to the historical operation data; calculating the object recommendation index of each of the existing objects according to the weight coefficient of each type of operation in the multiple types of operations and multiple co-occurrence matrices; generating an existing object recommendation list according to the multiple object recommendation indices, and sending the existing object recommendation list to the user terminal of the target user.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of computer technology, and in particular, to an object recommendation method and related devices. Background Art

[0002] With the development of the Internet and big data, many new service models have emerged in the field of venture capital, among which venture capital platforms are one of the effective ways to eliminate the "information gap" in the capital market. With the help of venture capital platforms, both parties of investment and financing can exchange information efficiently, and the operators of venture capital platforms can also take effective measures to improve the efficiency of docking between the two parties and provide valuable services to the market.

[0003] With the expansion of venture capital platforms and the sharp increase in the number of financing projects, investors are also facing higher and higher information screening costs when using venture capital platforms. Currently, there is no project recommendation method for scenarios such as equity financing platforms recommending financing projects to investors. Summary of the invention

[0004] The embodiments of the present application provide an object recommendation method and related devices for generating an object recommendation list based on user historical operation data.

[0005] A first aspect of an embodiment of the present application provides an object recommendation method, comprising:

[0006] Acquire preset target operations and historical operation data of target users, wherein each piece of the historical operation data is an operation performed by the target user on a stock object, the historical operation data includes multiple types of operations and multiple stock objects, and the preset target operation is one type of operation among the multiple types of operations;

[0007] Determine, according to the historical operation data, the co-occurrence matrix of the preset target operation and each type of operation in the historical operation data, wherein each element in any co-occurrence matrix represents a co-occurrence relationship between any two of the stock objects;

[0008] Calculating a weight coefficient of each type of operation in the multiple types of operations according to the historical operation data;

[0009] Calculating an object recommendation index for each of the stock objects according to a weight coefficient of each type of operation in the multiple types of operations and a plurality of the co-occurrence matrices;

[0010] Generate an existing object recommendation list according to the plurality of object recommendation indexes, and send the existing object recommendation list to a user terminal of the target user.

[0011] In a specific implementation, the determining, according to the historical operation data, respectively the co-occurrence matrix of the preset target operation and each type of operation in the historical operation data includes:

[0012] According to the number of operations of different operations performed by the target user on each stock object in the historical operation data, an operation matrix of each type of operation and different stock objects is established respectively;

[0013] According to the operation matrices of each type of operation and different stock objects, a co-occurrence matrix of the operation matrix corresponding to the preset target operation and each of the operation matrices is established respectively.

[0014] In a specific implementation, after determining the co-occurrence matrix of the preset target operation and each type of operation in the historical operation data respectively according to the historical operation data, the method further includes:

[0015] Standardizing the multiple co-occurrence matrices respectively to obtain corresponding multiple standard co-occurrence matrices;

[0016] The calculating the object recommendation index of each of the stock objects according to the weight coefficient of each type of operation in the multiple types of operations and the multiple co-occurrence matrices includes:

[0017] An object recommendation index of each of the stock objects is calculated according to a weight coefficient of each type of operation in the multiple types of operations and a plurality of the standard co-occurrence matrices.

[0018] In a specific implementation, the method further includes:

[0019] Acquire a plurality of pre-screened standard object tags and a standard object tag included in each of the stock objects;

[0020] Determine a label vector according to the number of occurrences of different standard object labels in the preset target operation of the historical operation data, and determine object vectors of different stock objects according to the number of occurrences of each of the standard object labels in different stock objects in the historical operation data;

[0021] Calculating correction indexes of different stock objects according to the label vector and the plurality of object vectors;

[0022] The calculating the object recommendation index of each of the stock objects according to the weight coefficient of each type of operation in the multiple types of operations and the multiple co-occurrence matrices includes:

[0023] The object recommendation index of each of the stock objects is calculated according to the weight coefficient of each type of operation in the multiple types of operations, the multiple co-occurrence matrices and the multiple correction indexes.

[0024] In a specific implementation, before acquiring a plurality of pre-screened standard object tags, the method further includes:

[0025] Acquire a preset object label of each of the stock objects and an object status of each of the stock objects;

[0026] Calculating the correlation between each preset object tag and the object status according to the plurality of preset object tags and the object status of the plurality of stock objects;

[0027] The standard object label is determined according to the correlation between each preset object label and the object condition.

[0028] In a specific implementation, the calculating the weight coefficient of each type of operation in the multiple types of operations according to the historical operation data includes:

[0029] The weight coefficient of each type of operation in the multiple types of operations is calculated according to the operation matrix of each type of operation and different stock objects in the historical data.

[0030] In a specific implementation, the method further includes:

[0031] Determining operation history sequences of different operations according to the number of operations of each type performed by the target user on different stock objects in the historical operation data;

[0032] The calculating the object recommendation index of each of the stock objects according to the weight coefficient of each type of operation in the multiple types of operations and the multiple co-occurrence matrices includes:

[0033] The object recommendation index of each of the stock objects is calculated according to the weight coefficient of each type of operation in the multiple types of operations, the multiple co-occurrence matrices and the operation history sequences of different operations.

[0034] A second aspect of an embodiment of the present application provides an object recommendation device, including:

[0035] an acquisition unit, configured to acquire preset target operations and historical operation data of a target user, wherein each piece of the historical operation data is an operation performed by the target user on a stock object, the historical operation data includes multiple types of operations and multiple stock objects, and the preset target operation is one type of operation among the multiple types of operations;

[0036] A determination unit, configured to determine, according to the historical operation data, a co-occurrence matrix of the preset target operation and each type of operation in the historical operation data, wherein each element in any co-occurrence matrix represents a co-occurrence relationship between any two of the stock objects;

[0037] A calculation unit, configured to calculate a weight coefficient of each type of operation in the plurality of types of operations according to the historical operation data;

[0038] The calculation unit is further used to calculate the object recommendation index of each of the stock objects according to the weight coefficient of each type of operation in the multiple types of operations and the multiple co-occurrence matrices;

[0039] A generating unit is used to generate a stock object recommendation list according to the plurality of object recommendation indexes, and send the stock object recommendation list to a user terminal of the target user.

[0040] In a specific implementation, the determining unit is specifically configured to establish an operation matrix of each type of operation and different stock objects according to the number of operations of different operations performed by the target user on each stock object in the historical operation data;

[0041] According to the operation matrices of each type of operation and different stock objects, a co-occurrence matrix of the operation matrix corresponding to the preset target operation and each of the operation matrices is established respectively.

[0042] In a specific implementation, the device further includes a processing unit;

[0043] The processing unit is used to perform standardization processing on the multiple co-occurrence matrices respectively to obtain corresponding multiple standard co-occurrence matrices;

[0044] The calculation unit is specifically used to calculate the object recommendation index of each of the stock objects according to the weight coefficient of each type of operation in the multiple types of operations and the multiple standard co-occurrence matrices.

[0045] In a specific implementation, the acquisition unit is specifically used to acquire a plurality of pre-screened standard object tags and a standard object tag included in each of the stock objects;

[0046] The determining unit is further configured to determine a label vector according to the number of occurrences of different standard object labels in the preset target operation of the historical operation data, and determine object vectors of different stock objects according to the number of occurrences of each of the standard object labels in different stock objects in the historical operation data;

[0047] The calculation unit is further used to calculate the correction index of different stock objects according to the label vector and the plurality of object vectors;

[0048] The calculation unit is specifically used to calculate the object recommendation index of each of the stock objects according to the weight coefficient of each type of operation in the multiple types of operations, the multiple co-occurrence matrices and the multiple correction indexes.

[0049] In a specific implementation, the acquisition unit is further used to acquire a preset object tag of each of the stock objects and an object status of each of the stock objects;

[0050] The calculation unit is further used to calculate the correlation between each preset object tag and the object status according to the plurality of preset object tags and the object status of the plurality of stock objects;

[0051] The determining unit is further configured to determine a standard object label according to the correlation between each preset object label and the object status.

[0052] In a specific implementation, the calculation unit is specifically used to calculate the weight coefficient of each type of operation in the multiple types of operations according to the operation matrix of each type of operation and different stock objects in the historical data.

[0053] In a specific implementation, the determining unit is further configured to determine an operation history sequence of different operations according to the number of operations of each type performed by the target user on different stock objects in the historical operation data;

[0054] The calculation unit is specifically used to calculate the object recommendation index of each of the stock objects according to the weight coefficient of each type of operation in the multiple types of operations, the multiple co-occurrence matrices and the operation history sequences of different operations.

[0055] A third aspect of an embodiment of the present application provides an object recommendation device, including:

[0056] CPU, memory and input / output interface;

[0057] The memory is a short-term storage memory or a persistent storage memory;

[0058] The central processing unit is configured to communicate with the memory and execute instructions in the memory to perform the method described in the first aspect.

[0059] A fourth aspect of the embodiments of the present application provides a computer program product comprising instructions, and when the computer program product is run on a computer, the computer is caused to execute the method described in the first aspect.

[0060] A fifth aspect of an embodiment of the present application provides a computer storage medium, wherein the computer storage medium stores instructions, and when the instructions are executed on a computer, the computer executes the method described in the first aspect.

[0061] It can be seen from the above technical scheme that the embodiments of the present application have the following advantages: the co-occurrence matrix of the preset target operation and each type of operation can be obtained according to the historical operation data of the target user, and the object recommendation index of each existing object can be calculated according to multiple co-occurrence matrices and weight coefficients of different operations, and an object recommendation list is generated and sent to the user terminal of the target user. The user can use the object recommendation list as a reference to select existing objects, thereby reducing the user's information screening cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 A system physical architecture diagram of the object recommendation method disclosed in an embodiment of the present application;

[0063] Figure 2 A system logic architecture diagram of the object recommendation method disclosed in the embodiment of the present application;

[0064] Figure 3 A schematic diagram of a flow chart of the object recommendation method disclosed in an embodiment of the present application;

[0065] Figure 4 A schematic diagram of the structure of the object recommendation device disclosed in the embodiment of the present application;

[0066] Figure 5 This is another structural schematic diagram of the object recommendation device disclosed in an embodiment of the present application. DETAILED DESCRIPTION

[0067] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0068] With the development of the Internet and big data, many new service models have emerged in the field of venture capital, among which venture capital platforms are one of the effective ways to eliminate the "information gap" in the capital market. With the help of venture capital platforms, both parties of investment and financing can exchange information efficiently, and the operators of venture capital platforms can also take effective measures to improve the efficiency of docking between the two parties and provide valuable services to the market.

[0069] As the scale of venture capital platforms expands and the number of financing projects increases dramatically, investors are also facing higher and higher information screening costs when using venture capital platforms. One way to solve this problem is intelligent recommendation technology.

[0070] Although intelligent recommendation technology has been applied very maturely in many Internet venture capital platforms, the existing intelligent recommendation algorithms are not suitable for scenarios such as equity financing venture capital platforms recommending financing projects to investors.

[0071] In order to adapt to the scenario of equity financing project screening, the embodiment of the present application provides an object recommendation method and related devices for generating an object recommendation list based on user historical operation data. It can recommend projects of interest to investors (i.e. users) based on their preferences, reduce the cost of project screening for investors and users, and improve docking efficiency.

[0072] The application scenario of the embodiment of the present application may include a server and a user terminal, wherein the server is used to generate an object recommendation list based on the user's historical operation data, and the user terminal is used to receive the object recommendation list sent by the server. The specific form of the server and the client is not limited here. For example, the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server (such as a cloud server provided by a cloud platform) that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The user terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited to this. The user terminal and the server can be directly or indirectly connected via wired or wireless communication, which is not limited in the embodiment of the present application.

[0073] See also Figure 1 In a specific embodiment, the application scenario of the embodiment of the present application may include a control server, a user terminal, a data source, a data cache, a computing engine, and a computing result storage. The control server is used to control the data source, the data cache, the computing engine, and the computing result storage to generate an object recommendation list and send the object recommendation list to the user terminal.

[0074] Specifically, the data source is used to provide the user's historical operation data to the control server, and then the control server sends the historical operation data to the calculation engine for calculation and returns the calculation result to the control server, and the control server sends the calculation result to the calculation result storage and returns the calculation result to the user terminal. In the calculation process, the control server will interact with the data cache to cache part of the data.

[0075] It can be known that, in practical applications, the computing engine may be a Spark cluster or a Hadoop cluster; the computing result storage may be an ES cluster, and the data cache may be MongoDB, but this is not specifically limited in the embodiments of the present application.

[0076] See also Figure 2In another specific embodiment, a system logic architecture disclosed in the embodiment of the present application includes: a co-occurrence matrix calculation module, an operation weight calculation module and a preliminary score calculation module. Among them, the co-occurrence matrix calculation module is used to determine the co-occurrence matrix of the preset target operation and each type of operation in the historical operation data and the preset target operation respectively according to the historical operation data; the operation weight calculation module is used to calculate the weight coefficient of each type of operation in the multiple types of operations in the historical operation data according to the historical operation data; the preliminary score calculation module is used to calculate the object recommendation index of each stock object according to the weight coefficient of each type of operation in the multiple types of operations and multiple co-occurrence matrices, and finally generate an object recommendation list.

[0077] In some specific embodiments, the system logic architecture of the embodiment of the present application also includes: a score correction module and a label filtering module; the co-occurrence matrix calculation module is also used to standardize the co-occurrence matrix of each type of operation and the target preset operation to eliminate the influence of overheated objects on the remaining co-occurrence relationships. Among them, the label filtering module is used to filter out standard object labels from multiple preset object labels; the score correction module is used to calculate the correction coefficient of each stock object based on the standard object label and historical operation data, and then correct the object recommendation index of different stock objects calculated by the preliminary score calculation matrix through the correction coefficient, and finally generate an object recommendation list.

[0078] The above describes the system logical architecture and system physical architecture in the embodiment of the present application. Figure 3 , the object recommendation method disclosed in the embodiment of the present application includes:

[0079] 301. Obtain historical operation data of a preset target operation and a target user, wherein each piece of historical operation data is an operation performed by the target user on an existing object, the historical operation data includes multiple types of operations and multiple existing objects, and the preset target operation is one type of operation among the multiple types of operations.

[0080] When the platform needs to recommend objects to target users, it first needs to obtain the target user's historical operation data on stock objects. Stock objects refer to all objects that can currently be recommended to the target user. In addition, each piece of historical operation data is an operation performed by the target user on a stock object, and the target user's historical operation data includes multiple pieces of historical operation data, multiple types of operations, and multiple stock objects. The aforementioned preset target operation is a type of operation included in the target user's historical data, which is pre-set by the user or platform developer based on experience and has the greatest impact on the recommendation index. All types of operations in the historical operation data except the preset target operation are secondary operations.

[0081] 302. Determine the co-occurrence matrix of the preset target operation and each type of operation in the historical operation data respectively according to the historical operation data, wherein each element in the co-occurrence matrix represents a co-occurrence relationship between any two stock objects.

[0082] According to the historical operation data of the target user obtained in step 301, the co-occurrence matrix of the preset target operation and each type of operation in the historical operation data is determined respectively. Among them, a co-occurrence matrix includes the number of times the target user performs a type of operation on different stock objects, and each element in the co-occurrence matrix describes the co-occurrence relationship between any two stock objects in the type of operation.

[0083] In some specific embodiments, it is also possible to establish an operation matrix for each type of operation and different stock objects based on the number of operations performed by the target user on each stock object in the historical operation data; then, based on the operation matrix for each type of operation and different stock objects, a co-occurrence matrix of the operation matrix corresponding to the preset target operation and each operation matrix is ​​established.

[0084] Specifically, an operation matrix of different operations may be established first, wherein an operation matrix specifically includes the number of times the target user performs operations on each stock object, wherein the type of operation performed is the operation matrix of that type of operation. Then, the transpose of the operation matrix of the preset target operation is multiplied by each operation matrix to obtain a co-occurrence matrix of the preset target operation and the preset target operation, and a co-occurrence matrix of the preset target operation and each secondary operation in step 301.

[0085] 303. Calculate the weight coefficient of each type of operation among multiple types of operations based on historical operation data.

[0086] Because different operations have different influences on the recommendation index, the weights of different operations need to be controlled by weight coefficients. In a specific embodiment, the weight coefficient of each type of operation in multiple types of operations is calculated based on the operation matrix of each type of operation and different stock objects in the historical data.

[0087] 304. Calculate an object recommendation index of each stock object according to a weight coefficient of each type of operation in the multiple types of operations and multiple co-occurrence matrices.

[0088] According to the multiple co-occurrence matrices determined in step 302 and the weight coefficients of each type of operation calculated in step 303, a recommendation index of each stock object, ie, an object recommendation index, can be calculated.

[0089] In some specific embodiments, before calculating the object recommendation index, it is also necessary to calculate the operation history sequence of different operations, where the operation history sequence is the number of times the target user performs a certain operation on each stock object. Then, the sum of the products of each operation, the operation co-occurrence matrix of the corresponding operation, and the operation history sequence of the corresponding operation is the object recommendation index of the target user for each stock object.

[0090] 305 . Generate a stock object recommendation list according to the multiple object recommendation indexes, and send the stock object recommendation list to a user terminal of the target user.

[0091] Finally, a stock object recommendation list of the target user is generated according to the object recommendation index of each stock object calculated in step 304, and the stock object recommendation list of the target user is sent to the user terminal of the target user.

[0092] In some specific embodiments, different stock objects may be arranged from high to low according to their object recommendation indexes, and then an object recommendation list with the object recommendation indexes from high to low is generated and sent to the user terminal of the target user.

[0093] In an embodiment of the present application, a co-occurrence matrix of a preset target operation and each type of operation can be obtained based on the historical operation data of the target user, and an object recommendation index of each existing object can be calculated based on multiple co-occurrence matrices and weight coefficients of different operations, and an object recommendation list can be generated and sent to the user terminal of the target user. The user can then use the object recommendation list as a reference to select existing objects, thereby reducing the user's information screening cost.

[0094] On the basis of the aforementioned steps 301 to 305, the object recommendation method of the embodiment of the present application further includes: obtaining a preset object label for each stock object and an object status for each stock object; calculating the correlation between each preset object label and the object status based on multiple preset object labels and the object status of multiple stock objects; determining a standard object label based on the correlation between each preset object label and the object status. Obtain multiple pre-screened standard object labels and standard object labels contained in each stock object; determine a label vector based on the number of occurrences of different standard object labels in preset target operations of historical operation data, and determine object vectors of different stock objects based on the number of occurrences of each standard object label in different stock objects in the historical operation data; calculate correction indexes for different stock objects based on the label vector and multiple object vectors;

[0095] Considering that the standard object labels of different stock objects will affect the operations of target users and thus affect the object recommendation index, it is necessary to correct the object recommendation index of different stock objects according to the object label of each stock object.

[0096] Specifically, it is first necessary to determine that all object labels of all stock objects are preset object labels, and then perform a correlation analysis on the preset object label of each stock object and the object status of the stock object. The preset object label that has a significant correlation with the object status is the standard object label, and the standard object label is used to calculate the correction index of each stock object.

[0097] It is known that the correlation analysis may be a binary logistic regression analysis or other correlation analysis method that can determine a preset object label that is significantly correlated with the object condition, and is not specifically limited here.

[0098] In some specific embodiments, first, a label vector is determined according to the number of occurrences of different standard object labels in the preset target operation of the historical operation data, and an object vector of different stock objects is determined according to the number of occurrences of each standard object label in different stock objects in the historical operation data; then, the correction index of different stock objects is calculated according to the intersection and union ratio of the label vector and multiple object vectors. Among them, the label vector refers to the number of occurrences of different labels in the operation history sequence of the preset target operation, and the object vector refers to the number of occurrences of different labels in a stock object.

[0099] Finally, the object recommendation index calculated in the aforementioned step 304 can be considered as a preliminary object recommendation index, and then the object recommendation index of each existing object can be multiplied by the correction index of the corresponding existing object to obtain the object recommendation index of the existing object for the target user, and the object recommendation list described in the aforementioned step 305 can be generated based on the final object recommendation index.

[0100] In another specific embodiment, the co-occurrence matrix calculated in the aforementioned step 302 may also be standardized to eliminate the influence of some popular stock objects of the platform on the target user's operation, thereby affecting the influence on the co-occurrence relationship in the co-occurrence matrix. Specifically, log likelihood ratio (LLR) processing, or cosine similarity, Pearson correlation coefficient algorithm, Euclidean distance algorithm, etc. may be used, which are not limited here.

[0101] In the embodiment of the present application, the influence of the popular stock objects of the platform on the target user's operation can be eliminated through standardization processing, thereby obtaining a more accurate co-occurrence matrix for calculating the object recommendation index. At the same time, the correction index of different stock objects is calculated according to the standard object label and the object recommendation index of different stock objects is corrected according to the correction index to obtain a more accurate object recommendation index.

[0102] The foregoing describes various implementation methods of the object recommendation method in the embodiments of the present application. The following describes a specific implementation method of the object recommendation method in the embodiments of the present application in the equity financing project screening scenario.

[0103] In order to provide personalized recommendations of projects (existing objects) of interest to investors (target users) based on their preferences, reduce the cost of project screening for investors and users, and improve docking efficiency.

[0104] First, the preset object label of each stock project is obtained from the investor's historical operation data, and the standard object label is filtered out according to the financing situation (object status) of the stock project.

[0105] Specifically, the labeling system can be constructed based on the following two principles: 1. Labels should cover as many indicators as possible that investors are concerned about when examining projects; 2. There should be sufficient correlation between project labels and investment behavior.

[0106] Based on the above principles, we first screened out four categories of first-level tags, namely: team tags, financial tags, products / technology / services, and market / environment, and further refined and derived a total of 44 second-level tags (preset object tags) including founder education, proportion of scientific researchers, annual turnover, annual net profit, etc. These tags cover various project characteristics that most investment institutions on the platform are concerned about.

[0107] On this basis, it is necessary to further use data to screen the above 44 secondary tags. Starting from the ultimate service goal of the venture capital platform, logistic regression is used to analyze the correlation between the historical investment and financing data of existing projects and the secondary tags. Specifically, all existing projects on the platform can be classified into two categories based on whether they have successfully raised funds, with successful financing being 1 and otherwise being 0. Binary logistic regression is used to analyze the correlation between the above 44 secondary tags and this classification result. In this example, the significance level in the binary logistic regression analysis result can be taken as 0.1, so all secondary indicators with a significance level less than 0.1 can be considered significant and can be used as labels to describe the project.

[0108] Then, the recommendation score (object recommendation index) is calculated based on the item co-occurrence relationship.

[0109] Specifically, we first build a corresponding co-occurrence matrix based on the user's historical operation records (historical operation data), then use the co-occurrence matrix combined with the target user's historical operation data to preliminarily calculate the recommendation scores of all existing items, and finally correct the recommendation scores based on the similarity (correction index) between each existing item and the target user obtained through the user-item similarity analysis, and sort the recommendation scores from high to low to obtain the final recommendation list for the target user.

[0110] In actual applications, the historical operation data of the target user is first collected and the main operations (preset target operations) are obtained.

[0111] If the target user's historical operation data includes click, favorite, and private message operations, you need to define a primary operation, that is, the operation that has the greatest impact on the recommendation results, and the rest of the operations are considered secondary operations. Click is defined as the primary operation here. Next, we will build a user operation matrix for different operations.

[0112] Define a set of items T = {T1, T2, ..., T n}, where n is the number of existing items on the platform. For click operations, define its click operation matrix C = (c m，n ), where c m，n For user U m Click on Project T n The number of times, m is the number of users in the platform. Define the transpose of the click operation matrix as CT, then the click-click co-occurrence matrix for click operations is C T *C, each element in the matrix describes the co-occurrence relationship between any two existing items. Further, the above co-occurrence matrix is ​​processed by LLR to eliminate the influence of overly popular items on other co-occurrence relationships. Finally, the standard click-click co-occurrence matrix is ​​obtained, which we record as LLR (C T *C).

[0113] Then, we construct a standard click-favorite co-occurrence matrix. First, we define the favorite operation matrix F = (f m，n ), where f m，n For user U m Favorite Project n The click-collection co-occurrence matrix for the collection operation is C T *F, the standard click-favorite co-occurrence matrix after LLR processing is LLR(C T *F).

[0114] Similarly, the private message operation matrix is ​​defined as S = (s m，n ), where s m，n For user U m Favorite Project n The standard click-private message co-occurrence matrix is ​​LLR(C T *S).

[0115] Finally, the preliminary project recommendation scores are calculated based on the standard click-click co-occurrence matrix, the standard click-collection co-occurrence matrix, and the standard click-private message co-occurrence matrix.

[0116] Specifically, for user U m , define its click operation history sequence HC = (hc n ), where hc n For user U m Click on Project Tn , note that HC is a row vector. Similarly, user U m The collection operation history sequence and private message operation history sequence are HF = (hf n ) and HS=(hs n ). Then, for user U m The project recommendation scoring formula is:

[0117] R=α*[LLR(C T *C)]*HCT+β*[LLR(C T *F)]*HF T +γ*[LLR(C T *S)]*HST (1)

[0118] In the above formula, R is all items about user U m The initial recommendation score is α, β and γ, which are the weight coefficients of click operation, collection operation and private message operation, respectively, reflecting the difference in contribution of different types of user operations to the project recommendation score. The specific calculation method is as follows:

[0119]

[0120]

[0121]

[0122] Finally, the aforementioned recommendation scores are modified according to the user-item similarity to obtain the final project recommendation scores for different stock projects.

[0123] Specifically, the aforementioned sequence R needs to be corrected according to the screened secondary tags.

[0124] If there are i secondary tags after screening, for user U m , count the number of times all tags appear in the click operation history sequence HC, and define the vector tag m =[j1, j2, ..., j i ], where t i is the number of times the i-th label appears in HC. n , define the vector item n =[k1, k2, ..., k i ], where k i is the i-th label in item T n The number of times it appears in .

[0125] For each item, the Jaccard correlation coefficient is used to evaluate the similarity between tagm and itemn, and the formula is:

[0126]

[0127] Then, for user U m , the similarity sequence of all items is:

[0128] jaccard m =[jaccard(tag m ,item1), jaccard(tag m ,item2),...,jaccard(tag m , item n )]

[0129] R and jaccard m Multiply the corresponding elements, and the resulting vector is the vector for user U m The recommendation scores of all items are sorted from high to low according to the scores, and the top N items are recommended to the user, which is the recommended item list of the user's best top N recommended items.

[0130] It can be known that, although the aforementioned scenario is to recommend items to target users, in actual applications, the stock objects in the embodiments of the present application may also be tasks or accounts, which are not specifically limited here.

[0131] The above describes some specific implementation methods of the object recommendation method in the embodiment of the present application in combination with specific scenarios. Figure 4 , the present application embodiment provides an object recommendation device, including:

[0132] The acquisition unit 401 is used to acquire the preset target operation and the historical operation data of the target user, wherein each piece of historical operation data is an operation performed by the target user on a stock object, the historical operation data includes multiple types of operations and multiple stock objects, and the preset target operation is one type of operation among the multiple types of operations;

[0133] A determination unit 402 is used to determine the co-occurrence matrix of the preset target operation and each type of operation in the historical operation data respectively according to the historical operation data, wherein each element in any co-occurrence matrix represents a co-occurrence relationship between any two stock objects;

[0134] A calculation unit 403, used to calculate a weight coefficient of each type of operation in the multiple types of operations according to the historical operation data;

[0135] The calculation unit 403 is further used to calculate the object recommendation index of each stock object according to the weight coefficient of each type of operation in the multiple types of operations and the multiple co-occurrence matrices;

[0136] The generating unit 404 is configured to generate a recommended list of existing objects according to a plurality of object recommendation indexes, and send the recommended list of existing objects to a user terminal of a target user.

[0137] In a specific implementation, the determination unit 402 is specifically configured to establish an operation matrix of each type of operation and different stock objects according to the number of operations performed by the target user on each stock object in the historical operation data;

[0138] According to the operation matrix of each type of operation and different stock objects, the operation matrix corresponding to the preset target operation and the co-occurrence matrix of each operation matrix are established respectively.

[0139] In a specific implementation, the apparatus further includes a processing unit;

[0140] A processing unit, used to perform standardization processing on the multiple co-occurrence matrices respectively to obtain corresponding multiple standard co-occurrence matrices;

[0141] The calculation unit 403 is specifically configured to calculate the object recommendation index of each stock object according to the weight coefficient of each type of operation in the multiple types of operations and multiple standard co-occurrence matrices.

[0142] In a specific implementation, the acquisition unit 401 is specifically configured to acquire a plurality of pre-screened standard object tags and a standard object tag included in each stock object;

[0143] The determining unit 402 is further configured to determine a label vector according to the number of occurrences of different standard object labels in the preset target operation of the historical operation data, and determine object vectors of different stock objects according to the number of occurrences of each standard object label in different stock objects in the historical operation data;

[0144] The calculation unit 403 is further used to calculate the correction index of different stock objects according to the label vector and the multiple object vectors;

[0145] The calculation unit 403 is specifically configured to calculate the object recommendation index of each stock object according to the weight coefficient of each type of operation in the multiple types of operations, multiple co-occurrence matrices and multiple correction indexes.

[0146] In a specific implementation, the acquisition unit 401 is further configured to acquire a preset object tag of each stock object and an object status of each stock object;

[0147] The calculation unit 403 is further used to calculate the correlation between each preset object tag and the object status according to the multiple preset object tags and the object status of the multiple stock objects;

[0148] The determination unit 402 is further configured to determine a standard object label according to the correlation between each preset object label and the object status.

[0149] In a specific implementation, the calculation unit 403 is specifically configured to calculate a weight coefficient of each type of operation in the plurality of types of operations according to an operation matrix of each type of operation in the historical data and different stock objects.

[0150] In a specific implementation, the determination unit 402 is further configured to determine the operation history sequence of different operations according to the number of operations of each type performed by the target user on different stock objects in the historical operation data;

[0151] The calculation unit 403 is specifically configured to calculate the object recommendation index of each stock object according to the weight coefficient of each type of operation in the multiple types of operations, multiple co-occurrence matrices, and operation history sequences of different operations.

[0152] In the embodiment of the present application, the determination unit 402 can obtain the co-occurrence matrix of the preset target operation and each type of operation based on the historical operation data of the target user, and then the calculation unit 403 calculates the object recommendation index of each existing object based on multiple co-occurrence matrices and weight coefficients of different operations. Finally, the object recommendation list is generated by the generation unit 404 and sent to the user terminal of the target user. The user can use the object recommendation list as a reference to select the existing objects, thereby reducing the user's information screening cost.

[0153] Figure 5 is a schematic diagram of the structure of an object recommendation device provided in an embodiment of the present application. The object recommendation device 500 may include one or more central processing units (CPU) 501 and a memory 505. The memory 505 stores one or more application programs or data.

[0154] The memory 505 may be a volatile storage or a persistent storage. The program stored in the memory 505 may include one or more modules, each of which may include a series of instruction operations in the object recommendation device. Furthermore, the central processor 501 may be configured to communicate with the memory 505 and execute a series of instruction operations in the memory 505 on the object recommendation device 500.

[0155] The object recommendation device 500 may also include one or more power supplies 502, one or more wired or wireless network interfaces 503, one or more input and output interfaces 504, and / or one or more operating systems, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc.

[0156] The CPU 501 can execute the aforementioned Figures 1 to 4The operations performed by the object recommendation device in the illustrated embodiment will not be described in detail here.

[0157] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0158] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0159] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0160] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0161] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, read-only memory), random access memory (RAM, random access memory), disk or optical disk and other media that can store program code.

Claims

1. An object recommendation method, characterized in that: include: Acquire preset target operations and historical operation data of target users, wherein each piece of the historical operation data is an operation performed by the target user on a stock object, the historical operation data includes multiple types of operations and multiple stock objects, the preset target operation is one type of operation among the multiple types of operations, and the stock objects include projects, tasks, or accounts; Determine, according to the historical operation data, the co-occurrence matrix of the preset target operation and each type of operation in the historical operation data, wherein each element in any co-occurrence matrix represents a co-occurrence relationship between any two of the stock objects; Calculating a weight coefficient of each type of operation in the multiple types of operations according to the historical operation data; Calculating an object recommendation index for each of the stock objects according to a weight coefficient of each type of operation in the multiple types of operations and a plurality of the co-occurrence matrices; The method further comprises: Acquire a plurality of pre-screened standard object tags and a standard object tag included in each of the stock objects; Determine a label vector according to the number of occurrences of different standard object labels in the preset target operation of the historical operation data, and determine object vectors of different stock objects according to the number of occurrences of each of the standard object labels in different stock objects in the historical operation data; Calculating correction indexes of different stock objects according to the label vector and the plurality of object vectors; The calculating the object recommendation index of each of the stock objects according to the weight coefficient of each type of operation in the multiple types of operations and the multiple co-occurrence matrices includes: Calculating an object recommendation index for each of the stock objects according to a weight coefficient of each type of operation in the multiple types of operations, a plurality of the co-occurrence matrices, and a plurality of the correction indexes; Generate an existing object recommendation list according to the plurality of object recommendation indexes, and send the existing object recommendation list to a user terminal of the target user.

2. The method according to claim 1, characterized in that The determining, according to the historical operation data, respectively the co-occurrence matrix of the preset target operation and each type of operation in the historical operation data comprises: According to the number of operations of different operations performed by the target user on each stock object in the historical operation data, an operation matrix of each type of operation and different stock objects is established respectively; According to the operation matrices of each type of operation and different stock objects, a co-occurrence matrix of the operation matrix corresponding to the preset target operation and each of the operation matrices is established respectively.

3. The method according to claim 1, characterized in that After determining the co-occurrence matrix of the preset target operation and each type of operation in the historical operation data respectively according to the historical operation data, the method further includes: Standardizing the multiple co-occurrence matrices respectively to obtain corresponding multiple standard co-occurrence matrices; The calculating the object recommendation index of each of the stock objects according to the weight coefficient of each type of operation in the multiple types of operations and the multiple co-occurrence matrices includes: An object recommendation index of each of the stock objects is calculated according to a weight coefficient of each type of operation in the multiple types of operations and a plurality of the standard co-occurrence matrices.

4. The method according to claim 1, characterized in that: Before acquiring a plurality of pre-screened standard object labels, the method further includes: Acquire a preset object label of each of the stock objects and an object status of each of the stock objects; Calculating the correlation between each preset object tag and the object status according to the plurality of preset object tags and the object status of the plurality of stock objects; The standard object label is determined according to the correlation between each preset object label and the object condition.

5. The method according to claim 1, characterized in that Calculating the weight coefficient of each type of operation in the multiple types of operations according to the historical operation data includes: The weight coefficient of each type of operation in the multiple types of operations is calculated according to the operation matrix of each type of operation and different stock objects in the historical operation data.

6. The method according to claim 1, characterized in that The method further comprises: Determining operation history sequences of different operations according to the number of operations of each type performed by the target user on different stock objects in the historical operation data; The calculating the object recommendation index of each of the stock objects according to the weight coefficient of each type of operation in the multiple types of operations and the multiple co-occurrence matrices includes: The object recommendation index of each of the stock objects is calculated according to the weight coefficient of each type of operation in the multiple types of operations, the multiple co-occurrence matrices and the operation history sequences of different operations.

7. An object recommendation device, characterized in that: include: an acquisition unit, configured to acquire preset target operations and historical operation data of a target user, wherein each piece of the historical operation data is an operation performed by the target user on a stock object, the historical operation data includes multiple types of operations and multiple stock objects, the preset target operation is one type of operation among the multiple types of operations, and the stock objects include projects, tasks, or accounts; A determination unit, configured to determine, according to the historical operation data, a co-occurrence matrix of the preset target operation and each type of operation in the historical operation data, wherein each element in any co-occurrence matrix represents a co-occurrence relationship between any two of the stock objects; A calculation unit, configured to calculate a weight coefficient of each type of operation in the plurality of types of operations according to the historical operation data; The calculation unit is further used to calculate the object recommendation index of each of the stock objects according to the weight coefficient of each type of operation in the multiple types of operations and the multiple co-occurrence matrices; The calculation unit is specifically used to obtain a plurality of pre-screened standard object labels and a standard object label contained in each of the stock objects; determine a label vector according to the number of occurrences of different standard object labels in the preset target operation of the historical operation data, and determine object vectors of different stock objects according to the number of occurrences of each of the standard object labels in different stock objects in the historical operation data; calculate correction indexes of different stock objects according to the label vector and the plurality of object vectors; Calculating an object recommendation index for each of the stock objects according to a weight coefficient of each type of operation in the multiple types of operations, a plurality of the co-occurrence matrices, and a plurality of the correction indexes; A generating unit is used to generate a stock object recommendation list according to the plurality of object recommendation indexes, and send the stock object recommendation list to a user terminal of the target user.

8. An object recommendation device, characterized in that: include: CPU, memory and input / output interface; The memory is a short-term storage memory or a persistent storage memory; The central processing unit is configured to communicate with the memory and execute instructions in the memory to perform the method according to any one of claims 1 to 6.

9. A computer storage medium, characterized in that The computer storage medium stores instructions, and when the instructions are executed on a computer, the computer is caused to execute the method according to any one of claims 1 to 6.

Images