CN114881108A - Commodity recommendation recall method, device, equipment and storage medium - Google Patents
Commodity recommendation recall method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN114881108A CN114881108A CN202210325064.3A CN202210325064A CN114881108A CN 114881108 A CN114881108 A CN 114881108A CN 202210325064 A CN202210325064 A CN 202210325064A CN 114881108 A CN114881108 A CN 114881108A
- Authority
- CN
- China
- Prior art keywords
- commodity
- commodities
- user
- modal
- association diagram
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 33
- 239000013598 vector Substances 0.000 claims abstract description 68
- 238000010586 diagram Methods 0.000 claims abstract description 65
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 23
- 230000000694 effects Effects 0.000 claims abstract description 14
- 238000005295 random walk Methods 0.000 claims abstract description 14
- 230000006399 behavior Effects 0.000 claims description 55
- 238000004364 calculation method Methods 0.000 claims description 25
- 230000007704 transition Effects 0.000 claims description 17
- 238000004590 computer program Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 238000012546 transfer Methods 0.000 claims description 4
- 238000007418 data mining Methods 0.000 abstract description 2
- 230000008569 process Effects 0.000 description 7
- 230000005012 migration Effects 0.000 description 6
- 238000013508 migration Methods 0.000 description 6
- 230000009471 action Effects 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/29—Graphical models, e.g. Bayesian networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Accounting & Taxation (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application discloses a commodity recommendation recall method, which relates to the field of data mining and comprises the following steps: obtaining multi-modal characteristics of each commodity on the commercial platform and a click behavior sequence of each user; generating a commodity association diagram according to the obtained multi-modal characteristics and the click behavior sequence, and performing multiple wandering on the commodity association diagram to obtain a plurality of wandering sequences taking the commodity as nodes; inputting a plurality of walking sequences into a pre-trained word vector model, and outputting a word vector corresponding to each commodity; and calculating the similarity parameter of the commodities according to the word vector corresponding to each commodity, and taking the commodities with the similarity value ranked at the top N as candidate commodities of the recall pool, wherein N is an integer greater than 1. According to the scheme, on the basis of a classical random walk algorithm based on a user behavior sequence, the multi-mode characteristics of the commodity are considered, the Martian effect caused by various deviations can be relieved to a certain extent, and the fairness and diversity of commodity recommendation are improved.
Description
Technical Field
The present application relates to the field of data mining, and in particular, to a method, an apparatus, a device, and a storage medium for recalling a recommendation of a commodity.
Background
Most e-commerce companies and retail companies utilize massive data to set a search or recommendation system on a website to promote sales, but for a long time, the main concern of research on the recommendation system is to improve the accuracy of commodity recommendation, which results in that different products on an e-commerce platform are not exposed fairly in recommendation, a few popular products are over-recommended, and most other products or suppliers do not get due attention.
The main reason for this is that the conventional recommendation algorithm exposes according to a sample selected by a model, and then enters model training to further select a new exposure sample, the model learns continuously based on biased samples, so that the overall feedback loop is continuously affected by bias, the system selection surface becomes narrower and forms "Martian effect", if this bias is not relieved, the experience of different users on the platform will be negatively affected, and as time goes on, this negative effect will be further aggravated by the interaction between the user and the platform.
Disclosure of Invention
Aiming at the defects in the prior art, the application provides a recall method for commodity recommendation according to multi-modal characteristics of commodities and clicking behaviors of users.
In order to achieve the purpose, the following technical scheme is adopted in the application:
the commodity recommendation recall method comprises the following steps:
obtaining multi-mode characteristics of each commodity on the commercial platform and a click behavior sequence of each user, wherein the click behavior comprises the clicked commodity and the time for clicking the commodity;
generating a commodity association diagram according to the obtained multi-modal characteristics and the click behavior sequence, and performing multiple wandering on the commodity association diagram to obtain a plurality of wandering sequences taking commodities as nodes;
inputting a plurality of the walking sequences into a pre-trained word vector model, and outputting a word vector corresponding to each commodity;
and calculating the similarity parameter of the commodities according to the word vector corresponding to each commodity, and taking the commodities with the similarity value ranked at the top N as candidate commodities of the recall pool, wherein N is an integer greater than 1.
Preferably, the multi-modal features include text features, image features, price features, and tag features.
Preferably, the generating a commodity association diagram according to the obtained multi-modal features and the click behavior sequence, and performing multiple walks on the commodity association diagram to obtain a plurality of walking sequences with commodities as nodes includes:
generating a commodity time sequence association diagram containing the user click co-occurrence edges according to commodities clicked by each user and the time for clicking the commodities, and obtaining the weight of the user click behavior co-occurrence edges by utilizing a transfer probability calculation formula;
processing the multi-modal characteristics of each commodity according to a proximity algorithm to obtain k most similar commodities of each commodity, adding multi-modal vector edges between the current commodity and the k most similar commodities in the commodity time sequence association diagram to obtain a commodity association diagram, and determining the weight of the multi-modal vector edges, wherein k is an integer greater than or equal to 1;
and carrying out random walk on the commodity association graph according to the weight of the user click behavior co-occurrence edge to obtain a plurality of walk sequences taking the commodities as nodes.
Preferably, the obtaining the weight of the co-occurrence edge of the user click behaviors by using a transition probability calculation formula includes:
obtaining the activity of each user and the popularity of each commodity, and adding a user activity parameter and a commodity popularity parameter into a commodity similarity calculation formula to obtain an expression (1):
wherein E is ij Similarity of item i to item j, B i→j And | B i→j L represents the user set and the user number, U, of purchasing the commodity j after purchasing the commodity i respectively i 、U j Is the set of users who purchased goods i and j, | U i ∩U j I is the co-purchase of goods i anduser set of item j, q u For user liveness, p i And p j The popularity of each of the commodity i and the commodity j, and the value which is equal to the norm; t is t u，i Represents the time t of the user u for clicking the commodity i u，j Represents the time of the user u clicking the commodity j, | t u，i -t u，j L is the time interval between the user u clicking the commodity i and clicking the commodity j;
the obtained E ij Substituting the obtained product into a commodity transition probability calculation formula to obtain an expression (2):
where p (ji) is the probability of the product i moving to the product j, and n (i) is the set of all the related products of the product i.
Preferably, the processing the multi-modal features of each commodity according to the proximity algorithm to obtain k commodities most similar to each commodity, adding a multi-modal vector edge between the current commodity and the k commodities most similar to the current commodity in the commodity time-series association diagram to obtain a commodity association diagram, and determining a weight of the multi-modal vector edge, where k is an integer greater than or equal to 1, includes:
calculating cosine similarity between each feature of each commodity, and setting different weights for the cosine similarities corresponding to different features;
calculating the weighted cosine similarity among the commodities according to the cosine similarity corresponding to different characteristics of the commodities and the corresponding weight;
arranging the weighted cosine similarity among the commodities in a descending order to obtain k most similar commodities of each commodity, wherein k is an integer greater than or equal to 1;
and adding multi-modal vector edges between the current commodity and the k most similar commodities in the commodity time sequence association diagram to obtain a commodity association diagram, wherein the weighted cosine similarity between the commodities is the weight corresponding to the multi-modal vector edges.
Preferably, the randomly walking the commodity association graph according to the weight of the user click behavior co-occurrence edge to obtain a plurality of walking sequences using commodities as nodes, including: and carrying out random walk on the commodity association diagram according to the transition probability among the commodities to obtain a plurality of walk sequences taking the commodities as nodes, wherein each walk sequence comprises co-occurrence side information of user click behaviors and multi-mode vector side information.
A recall apparatus for merchandise recommendations comprising:
the system comprises an acquisition module, a display module and a display module, wherein the acquisition module is used for acquiring multi-modal characteristics of each commodity on an e-commerce platform and a click behavior sequence of each user, and the click behavior comprises a clicked commodity and time for clicking the commodity;
the calculation module is used for generating a commodity association diagram according to the obtained multi-modal characteristics and the click behavior sequence, and performing multiple wandering on the commodity association diagram to obtain a plurality of wandering sequences taking commodities as nodes;
the conversion module is used for inputting a plurality of the walking sequences into a pre-trained word vector model and outputting a word vector corresponding to each commodity;
and the selection module is used for calculating the similarity parameter of the commodity according to the word vector corresponding to each commodity, and taking the commodity with the similarity value ranked at the top N as a candidate commodity of the recall pool, wherein N is an integer greater than 1.
Preferably, the calculation module includes:
the generation unit is used for generating a commodity time sequence association diagram containing the user click co-occurrence edges according to the commodity clicked by each user and the time for clicking the commodity, and obtaining the weight of the user click behavior co-occurrence edges by using a transition probability calculation formula;
the adding unit is used for processing the multi-modal characteristics of each commodity according to a proximity algorithm to obtain k most similar commodities of each commodity, adding multi-modal vector edges between the current commodity and the k most similar commodities in the commodity time sequence association diagram to obtain a commodity association diagram, and determining the weight of the multi-modal vector edges, wherein k is an integer greater than or equal to 1;
and the walking unit is used for carrying out random walking on the commodity association graph according to the weight of the co-occurrence edge of the user clicking behaviors to obtain a plurality of walking sequences taking the commodities as nodes.
An electronic device comprising a memory and a processor, the memory storing one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement a method of recalling a recommendation for an item as claimed in any one of the preceding claims.
A computer-readable storage medium storing a computer program which, when executed, causes a computer to implement a method of recalling a recommendation of an item as in any one of the above.
The invention has the following beneficial effects:
according to the scheme, on the basis of a classical random walk algorithm based on a user behavior sequence, the multi-mode characteristics of the commodity are considered, the Martian effect caused by various deviations can be relieved to a certain extent, and the fairness and diversity of commodity recommendation are improved.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without inventive exercise.
Fig. 1 is a first flowchart of a method for recalling a recommendation of an article according to embodiment 1 of the present application;
fig. 2(a) is a commodity association diagram obtained based on the click behavior of the user U1 in embodiment 2 of the present application;
FIG. 2(b) is a schematic diagram of a walking sequence in example 2 of the present application;
FIG. 2(c) is a diagram of a vector transformation model in embodiment 2 of the present application;
FIG. 3 is a schematic diagram of a recall device for merchandise recommendation implemented in embodiment 3 of the present application;
FIG. 4 is a schematic diagram of a computing module of a recall device for merchandise recommendation implemented in embodiment 4 of the present application;
fig. 5 is a schematic view of an electronic device implementing a method for recalling a recommendation of an article according to embodiment 5 of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be described clearly and completely with reference to the accompanying drawings, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The terms "first," "second," and the like in the claims and in the description of the present application are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order, it being understood that the terms so used are interchangeable under appropriate circumstances and are merely used to describe a distinguishing manner between similar elements in the embodiments of the present application and that the terms "comprising" and "having" and any variations thereof are intended to cover a non-exclusive inclusion such that a process, method, system, article, or apparatus that comprises a list of elements is not necessarily limited to those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.
Example 1
As shown in fig. 1, a method for recalling recommendation of goods includes the following steps:
s110, obtaining multi-mode characteristics of each commodity on the E-commerce platform and a click behavior sequence of each user, wherein the click behavior comprises the clicked commodity and the time of clicking the commodity;
s120, generating a commodity association diagram according to the obtained multi-modal features and the click behavior sequence, and performing multiple wandering on the commodity association diagram to obtain a plurality of wandering sequences taking commodities as nodes;
s130, inputting a plurality of the walking sequences into a pre-trained word vector model, and outputting a word vector corresponding to each commodity;
s140, calculating similarity parameters of the commodities according to the word vectors corresponding to the commodities, and taking the commodities with the similarity value ranked at the top N as candidate commodities in a recall pool, wherein N is an integer larger than 1.
According to embodiment 1, in this embodiment, firstly, multi-modal features of all the commodities on the merchant recommendation platform are obtained, the multi-modal features include text features, image features, price features and tag features, the multi-modal features are represented by high-dimensional vectors, then click behavior sequences of all the users on the platform are obtained, the click behavior sequences include the commodities clicked by the users and the time of clicking the commodities, a time sequence association diagram of the commodities is constructed according to the relation between the commodities and the time in the click behavior sequences of the users, edges in the diagram are co-occurrence edges clicked by the users and represent the commodities clicked by the same user, the arrow direction of each edge is determined according to the time sequence, a time interval parameter, a commodity popularity parameter and a user activity parameter are added in a transition probability calculation formula of a conventional random walk algorithm, then the transition probability among the commodities is calculated to reduce the influence of selective deviation on a recommendation result, and the obtained inter-commodity probability is used as the weight of the user clicking the co-occurrence edge of the corresponding commodity, and then the cosine similarity between each feature of each commodity is calculated, such as calculating the cosine similarity of the text features of the commodity A and the commodity B, the cosine similarity of the image features, the cosine similarity of the price features and the cosine similarity of the label features respectively, wherein the calculation of the cosine similarity is only one of the methods, and can also be used for calculating Euclidean distance, or calculating Pearson correlation coefficient and the like, and the substitution can be carried out according to actual needs when in specific use, different weights are given to the cosine similarities calculated according to different features according to the importance of different features, for example, the weight of the cosine similarity of the text features of the commodity A and the commodity B is set to 0.2, the cosine similarity of the image features is set to 0.5, the cosine similarity of the price features is set to 0.1, and the cosine similarity of the label features is set to 0.2, calculating the sum of weighted cosine similarity of each characteristic to obtain weighted cosine similarity among commodities, determining k commodities which are most similar to each commodity by combining a proximity algorithm, namely knn algorithm, according to the weighted cosine similarity among the commodities, wherein k is an integer more than or equal to 1, establishing a connecting edge for each commodity and the k most similar commodities in a commodity time sequence association diagram, wherein the edge is called a multi-mode vector edge, the weight is the weighted cosine similarity of the commodity obtained by calculating the cosine similarity, the arrow point of each edge is the k most similar commodities of the current commodity calculated by using knn algorithm, then carrying out random walk in the commodity association diagram according to the transition probability among the commodities to obtain a plurality of walk sequences taking the commodities as the starting point, and each walk sequence comprises information of user clicking the co-occurrence edge and information of the multi-mode vector edge, and finally, calculating cosine similarity among commodities by using the word vector, and taking all commodities with the cosine similarity arranged at top N as a commodity candidate set of a recall pool, wherein N is an integer greater than 1. On the basis of a traditional random walk model based on a user behavior sequence, multi-mode characteristics (such as commodity text characteristics, picture characteristics, price characteristics, label characteristics and the like) and user click timing sequence characteristics of commodities are considered at the same time, time interval parameters, commodity popularity parameters and user activity parameters are added during calculation of transition probability among the commodities, influences caused by various deviations are reduced, Martian's effect caused by various deviations can be relieved to a certain extent, and fairness and diversity of recommendation are improved.
Example 2
As shown in fig. 2, a method for recalling a recommendation of an article includes:
s210, obtaining multi-mode characteristics of each commodity on the E-commerce platform and a click behavior sequence of each user, wherein the click behavior comprises the clicked commodity and the time of clicking the commodity;
s220, generating a commodity time sequence association diagram containing the user click co-occurrence edges according to commodities clicked by each user and the time for clicking the commodities, and obtaining the weight of the user click behavior co-occurrence edges by using a transfer probability calculation formula;
s230, processing the multi-modal characteristics of each commodity according to a proximity algorithm to obtain k most similar commodities of each commodity, adding multi-modal vector edges between the current commodity and the k most similar commodities in the commodity time sequence association diagram to obtain a commodity association diagram, and determining the weight of the multi-modal vector edges, wherein k is an integer greater than or equal to 1;
s240, carrying out random walk on the commodity association graph according to the weight of the user click behavior co-occurrence edge to obtain a plurality of walk sequences with commodities as nodes;
s250, inputting a plurality of the walking sequences into a pre-trained word vector model, and outputting a word vector corresponding to each commodity;
s260, calculating similarity parameters of the commodities according to the word vectors corresponding to the commodities, and taking the commodities with the similarity value ranked at the top N as candidate commodities in a recall pool, wherein N is an integer larger than 1.
According to the embodiment 2, in the embodiment, description text information, first page picture information, annual average transaction price information and first-level label code information of all commodities on the commodity recommendation platform are firstly required to be obtained, because the text information and the pictures can well express multi-mode characteristics of the commodities to serve for calculating the similarity between the commodities later, the first-level label code information of the commodities can accurately express the categories of the commodities to which the commodities belong, the accuracy rate of calculating the similarity of the commodities is provided, the text information and the picture information are trained through a bert model and a cnn algorithm respectively to generate a 128-dimensional real value vector to obtain the text characteristics and the image characteristics of the commodities, a commodity set C and a user set U on the platform are simultaneously obtained to obtain a click behavior sequence of the commodities clicked by the user within the last 30 days, and the click behaviors comprise the commodities clicked by the user and the time of clicking the commodities, then, a commodity time sequence association graph is constructed according to commodities clicked by users and time for clicking the commodities, the edges in the graph are co-occurrence edges clicked by the users and represent commodities clicked by the same user, the commodities are represented by solid lines, as shown in a graph (a) in FIG. 2, a commodity A is clicked after a commodity D is clicked by a user U1, a commodity B is clicked after a commodity C is clicked, a commodity A and a commodity E are clicked after a commodity B is clicked, a commodity F is clicked after a commodity E is clicked, the arrow direction of each edge is determined according to a time sequence, a time interval parameter, a commodity popularity parameter and a user activity parameter are added into a transition probability calculation formula of a traditional random walk algorithm, and then the transition probability among the commodities is calculated, wherein the specific calculation formula is as follows:
wherein E is ij Is the similarity of the item i to the item j, P (j | i) is the probability of the item i transitioning to the item j, B i→j And | B i→j L represents the user set and the user number, U, of purchasing the commodity j after purchasing the commodity i respectively i 、U j Is the set of users who purchased goods i and j, | U i ∩U j L is a set of users who jointly purchase item i and item j, q u For user liveness, p i And p j The popularity of each of the commodity i and the commodity j, which is a constant, t u，i Represents the time t of the user u for clicking the commodity i u，j Represents the time of the user u clicking the commodity j, | t u，i -t u，j L is the time interval between the user u clicking on item i and clicking on item j, and N (i) is the set of all associated items for item i.
According to the time interval parameter, the shorter the co-occurrence time between two commodities is, the larger the weight value of the user clicking the co-occurrence edge is; the action influence degree of the active users is reduced through the user activity degree parameters, and the fairness of the active users and the inactive users is improved, if the action frequency of the users is higher, the more active the users are; considering the popularity deviation caused by popular commodities, through the commodity popularity parameters, popularity punishment is introduced in the weight calculation of edges in the composition process, the weight of the popular commodities is reduced, so that the commodities with low popularity can be more organically explored during random walk, and the popularity deviation is reduced.
Then, calculating cosine similarity between each feature of each commodity, assigning different weights to the cosine similarity calculated according to different features according to importance degrees of different features, calculating the sum of weighted cosine similarity of each feature to obtain weighted cosine similarity between each commodity, determining k commodities most similar to each commodity according to the weighted cosine similarity between each commodity and knn algorithm, wherein k is an integer greater than or equal to 1, such as commodity B is similar to commodity A through knn algorithm, commodity C is similar to commodity E, commodity E is similar to commodity D, establishing a connecting edge in a commodity time sequence association diagram for each commodity and k commodities most similar to the commodity, the edge is called a multi-modal vector edge, as shown in diagram (a) in FIG. 2, a dotted line in the diagram is a multi-modal vector edge between commodities, and the weight is the weighted cosine similarity between corresponding commodities obtained through calculation, the arrow direction of each edge is k most similar commodities of the current commodity calculated by using knn algorithm, then random walk is performed in the commodity association map according to the transition probability among the commodities, so that a plurality of walk sequences with the commodities as the starting points can be obtained, as shown in a diagram (b) in fig. 2, each walk sequence comprises information of clicking a co-occurrence edge by a user and information of a multi-modal vector edge, finally all the walk sequences are input into a word vector model trained by using a word2vec algorithm, so that a word vector corresponding to each commodity can be obtained, that is, a high-dimensional vector is converted into a low-dimensional vector, as shown in a diagram (c) in fig. 2, each low-dimensional vector comprises commodity information and associated information between each low-dimensional vector and a related commodity, and finally, the cosine similarity among the commodities is calculated by using the word vector, as shown below:
a, B are word vectors corresponding to the commodities A and B, A i 、B i Respectively A, B.
And taking a set consisting of the commodities with cosine similarity values ranked at top N as a commodity candidate set of the recall pool, wherein N is an integer greater than 1. On the basis of a classical random walk algorithm based on a user behavior sequence, the multi-modal characteristics of the commodity are mainly considered, and the multi-modal characteristics of the commodity are irrelevant to the popularity of the commodity, the activeness of the user and selective deviation, so that the Martian effect caused by various deviations can be relieved to a certain extent, and the fairness and diversity of recommendation are improved.
Example 3
As shown in fig. 3, a recall apparatus for recommendation of an article includes:
the system comprises an acquisition module 10, a display module and a display module, wherein the acquisition module is used for acquiring multi-modal characteristics of each commodity on an e-commerce platform and a click behavior sequence of each user, and the click behavior comprises a clicked commodity and time for clicking the commodity;
the calculation module 20 is configured to generate a commodity association diagram according to the obtained multi-modal features and the click behavior sequence, and perform multiple walks on the commodity association diagram to obtain a plurality of walking sequences using commodities as nodes;
the conversion module 30 is configured to input a plurality of the walking sequences into a pre-trained word vector model, and output a word vector corresponding to each commodity;
and the selecting module 40 is configured to calculate a similarity parameter of each commodity according to the word vector corresponding to each commodity, and use the commodity with the similarity value ranked in the top N as a candidate commodity of the recall pool, where N is an integer greater than 1.
One embodiment of the above apparatus may be: the obtaining module 10 obtains a multi-modal feature of each commodity on the merchant platform and a click behavior sequence of each user, wherein the click behavior comprises a clicked commodity and time for clicking the commodity; the calculation module 20 generates a commodity association diagram according to the obtained multi-modal features and the click behavior sequence, and performs multiple wandering on the commodity association diagram to obtain a plurality of wandering sequences taking the commodity as nodes; the conversion module 30 inputs a plurality of the walking sequences into a pre-trained word vector model, and outputs a word vector corresponding to each commodity; the selecting module 40 calculates similarity parameters of the commodities according to the word vectors corresponding to each commodity, and takes the commodity with the similarity value ranked at the top N as a candidate commodity of the recall pool, where N is an integer greater than 1.
Example 4
As shown in fig. 4, a computing module 20 of a recall apparatus for recommendation of goods includes:
the generation unit 22 is configured to generate a commodity timing association diagram including a user click co-occurrence side according to a commodity clicked by each user and time for clicking the commodity, and obtain a weight of the user click behavior co-occurrence side by using a transition probability calculation formula;
the adding unit 24 is configured to process the multi-modal features of each commodity according to a proximity algorithm to obtain k most similar commodities for each commodity, add multi-modal vector edges between the current commodity and the k most similar commodities in the commodity time sequence association diagram to obtain a commodity association diagram, and determine a weight of the multi-modal vector edges, where k is an integer greater than or equal to 1;
and the migration unit 26 is configured to perform random migration on the commodity association graph according to the weight of the user click behavior co-occurrence edge, so as to obtain a plurality of migration sequences using the commodities as nodes.
One implementation of the computing module 20 of the above apparatus may be: the generating unit 22 generates a commodity time sequence association diagram containing the user click co-occurrence edges according to the commodity clicked by each user and the time for clicking the commodity, and obtains the weight of the user click behavior co-occurrence edges by using a transfer probability calculation formula; the adding unit 24 processes the multi-modal features of each commodity according to a proximity algorithm to obtain k most similar commodities of each commodity, adds multi-modal vector edges between the current commodity and the k most similar commodities in the commodity time sequence association diagram to obtain a commodity association diagram, and determines weights of the multi-modal vector edges, wherein k is an integer greater than or equal to 1; the migration unit 26 performs random migration on the commodity association graph according to the weight of the user click behavior co-occurrence edge to obtain a plurality of migration sequences using commodities as nodes.
Example 5
As shown in fig. 5, an electronic device comprises a memory 501 and a processor 502, wherein the memory 501 is used for storing one or more computer instructions, and the one or more computer instructions are executed by the processor 502 to implement the above-mentioned electronic contract-based goods transaction method.
It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working process of the electronic device described above may refer to the corresponding process in the foregoing method embodiment, and is not described herein again.
A computer-readable storage medium storing a computer program which, when executed, causes a computer to implement a method for trading goods based on electronic contracts as described above.
Illustratively, a computer program may be divided into one or more modules/units, one or more modules/units are stored in the memory 501 and executed by the processor 502, and the input interface 505 and the output interface 506 perform I/O interface transmission of data to complete the present invention, and one or more modules/units may be a series of computer program instruction segments for describing the execution of the computer program in a computer device, which can perform specific functions.
The computer device may be a desktop computer, a notebook, a palm computer, a cloud server, or other computing devices. The computer device may include, but is not limited to, the memory 501 and the processor 502, and those skilled in the art will appreciate that the present embodiment is only an example of the computer device, and does not constitute a limitation of the computer device, and may include more or less components, or combine some components, or different components, for example, the computer device may further include the input device 507, a network access device, a bus, and the like.
The Processor 502 may be a Central Processing Unit (CPU), other general purpose Processor 502, a Digital Signal Processor 502 (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic, discrete hardware components, etc. The general purpose processor 502 may be a microprocessor 502 or the processor 502 may be any conventional processor 502 or the like.
The storage 501 may be an internal storage unit of the computer device, such as a hard disk or a memory of the computer device. The memory 501 may also be an external storage device of the computer device, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), etc. provided on the computer device, further, the memory 501 may also include both an internal storage unit and an external storage device of the computer device, the memory 501 is used for storing computer programs and other programs and data required by the computer device, the memory 501 may also be used for temporarily storing in the output device 508, and the aforementioned storage Media include various Media capable of storing program codes, such as a usb disk, a removable hard disk, a ROM503, a RAM504, a disk, or an optical disk.
The above description is only an embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions within the technical scope of the present invention are intended to be covered by the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the appended claims.
Claims (10)
1. A method for recalling recommendations for merchandise, comprising the steps of:
obtaining multi-mode characteristics of each commodity on the commercial platform and a click behavior sequence of each user, wherein the click behavior comprises the clicked commodity and the time for clicking the commodity;
generating a commodity association diagram according to the obtained multi-modal characteristics and the click behavior sequence, and performing multiple wandering on the commodity association diagram to obtain a plurality of wandering sequences taking commodities as nodes;
inputting a plurality of the walking sequences into a pre-trained word vector model, and outputting a word vector corresponding to each commodity;
and calculating the similarity parameter of the commodities according to the word vector corresponding to each commodity, and taking the commodities with the similarity value ranked at the top N as candidate commodities of the recall pool, wherein N is an integer greater than 1.
2. The method of claim 1, wherein the multi-modal features comprise text features, image features, price features, and label features.
3. The method for recalling recommendation of goods according to claim 2, wherein said generating a goods association diagram according to the obtained multi-modal features and click behavior sequence, and making multiple walks on the goods association diagram to obtain a plurality of walking sequences with goods as nodes comprises:
generating a commodity time sequence association diagram containing the user click co-occurrence edges according to commodities clicked by each user and the time for clicking the commodities, and obtaining the weight of the user click behavior co-occurrence edges by utilizing a transfer probability calculation formula;
processing the multi-modal characteristics of each commodity according to a proximity algorithm to obtain k most similar commodities of each commodity, adding multi-modal vector edges between the current commodity and the k most similar commodities in the commodity time sequence association diagram to obtain a commodity association diagram, and determining the weight of the multi-modal vector edges, wherein k is an integer greater than or equal to 1;
and carrying out random walk on the commodity association graph according to the weight of the user click behavior co-occurrence edge to obtain a plurality of walk sequences taking the commodities as nodes.
4. The method as claimed in claim 3, wherein the obtaining the weight of the co-occurrence edge of the user click behaviors by using a transition probability calculation formula comprises:
obtaining the activity of each user and the popularity of each commodity, and adding a user activity parameter and a commodity popularity parameter into a commodity similarity calculation formula to obtain an expression (1):
wherein E is ij Similarity between item i and item j, B i→j And | B i→j L represents the user set and the user number, U, of purchasing the commodity j after purchasing the commodity i respectively i ，U j Is the set of users who purchased goods i and j, | U i ∩U j L is a set of users who buy the goods i and j together, q u For user liveness, p i And p j The popularity of each of the commodity i and the commodity j, and oc is a constant; t is t u，i Represents the time t of the user u for clicking the commodity i u，j Represents the time of the user u clicking the commodity j, | t u，i -t u，j L is the time interval between the user u clicking the commodity i and clicking the commodity j;
the obtained E ij Substituting the obtained product into a commodity transition probability calculation formula to obtain an expression (2):
where P (j | i) is the probability of the transition of item i to item j, and n (i) is the set of all associated items for item i.
5. The method as claimed in claim 3, wherein the processing the multi-modal features of each item according to the proximity algorithm to obtain k items most similar to each item, adding a multi-modal vector edge between the current item and the k items most similar to the current item in the item time-series correlation diagram to obtain an item correlation diagram, and determining the weight of the multi-modal vector edge, wherein k is an integer greater than or equal to 1, comprises:
calculating cosine similarity between each feature of each commodity, and setting different weights for the cosine similarities corresponding to different features;
calculating the weighted cosine similarity among the commodities according to the cosine similarity corresponding to different characteristics of the commodities and the corresponding weight;
arranging the weighted cosine similarity among the commodities in a descending order to obtain k most similar commodities of each commodity, wherein k is an integer greater than or equal to 1;
and adding multi-modal vector edges between the current commodity and the k most similar commodities in the commodity time sequence association diagram to obtain a commodity association diagram, wherein the weighted cosine similarity between the commodities is the weight corresponding to the multi-modal vector edges.
6. The method as claimed in claim 4, wherein the randomly walking the product association graph according to the weight of the co-occurrence edge of the user click behavior to obtain a plurality of walking sequences using the product as a node comprises: and carrying out random walk on the commodity association diagram according to the transition probability among the commodities to obtain a plurality of walk sequences taking the commodities as nodes, wherein each walk sequence comprises co-occurrence side information of user click behaviors and multi-mode vector side information.
7. A recall apparatus for merchandise recommendation, comprising:
the system comprises an acquisition module, a display module and a display module, wherein the acquisition module is used for acquiring multi-modal characteristics of each commodity on an e-commerce platform and a click behavior sequence of each user, and the click behavior comprises a clicked commodity and time for clicking the commodity;
the calculation module is used for generating a commodity association diagram according to the obtained multi-modal characteristics and the click behavior sequence, and performing multiple wandering on the commodity association diagram to obtain a plurality of wandering sequences taking commodities as nodes;
the conversion module is used for inputting a plurality of the walking sequences into a pre-trained word vector model and outputting a word vector corresponding to each commodity;
and the selection module is used for calculating the similarity parameter of the commodity according to the word vector corresponding to each commodity, and taking the commodity with the similarity value ranked at the top N as a candidate commodity of the recall pool, wherein N is an integer greater than 1.
8. The article recommendation recall device of claim 7, wherein the computing module comprises:
the generation unit is used for generating a commodity time sequence association diagram containing the user click co-occurrence edges according to the commodity clicked by each user and the time for clicking the commodity, and obtaining the weight of the user click behavior co-occurrence edges by using a transition probability calculation formula;
the adding unit is used for processing the multi-modal characteristics of each commodity according to a proximity algorithm to obtain k most similar commodities of each commodity, adding multi-modal vector edges between the current commodity and the k most similar commodities in the commodity time sequence association diagram to obtain a commodity association diagram, and determining the weight of the multi-modal vector edges, wherein k is an integer greater than or equal to 1;
and the walking unit is used for carrying out random walking on the commodity association graph according to the weight of the co-occurrence edge of the user clicking behaviors to obtain a plurality of walking sequences taking the commodities as nodes.
9. An electronic device comprising a memory and a processor, the memory configured to store one or more computer instructions, wherein the one or more computer instructions are executable by the processor to implement a method of recalling a recommendation for an item as recited in any one of claims 1-6.
10. A computer-readable storage medium storing a computer program, wherein the computer program is configured to cause a computer to execute a method of recalling a recommendation of an item according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210325064.3A CN114881108A (en) | 2022-03-30 | 2022-03-30 | Commodity recommendation recall method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210325064.3A CN114881108A (en) | 2022-03-30 | 2022-03-30 | Commodity recommendation recall method, device, equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114881108A true CN114881108A (en) | 2022-08-09 |
Family
ID=82668821
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210325064.3A Pending CN114881108A (en) | 2022-03-30 | 2022-03-30 | Commodity recommendation recall method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114881108A (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108898459A (en) * | 2018-06-25 | 2018-11-27 | 中国联合网络通信集团有限公司 | A kind of Method of Commodity Recommendation and device |
| US20220019739A1 (en) * | 2019-02-21 | 2022-01-20 | Beijing Jingdong Shangke Information Technology Co., Ltd. | Item Recall Method and System, Electronic Device and Readable Storage Medium |
-
2022
- 2022-03-30 CN CN202210325064.3A patent/CN114881108A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108898459A (en) * | 2018-06-25 | 2018-11-27 | 中国联合网络通信集团有限公司 | A kind of Method of Commodity Recommendation and device |
| US20220019739A1 (en) * | 2019-02-21 | 2022-01-20 | Beijing Jingdong Shangke Information Technology Co., Ltd. | Item Recall Method and System, Electronic Device and Readable Storage Medium |
Non-Patent Citations (2)
| Title |
|---|
| WANG JIZHE ET AL.: ""Billion-scale Commodity Embedding for E-commerce Recommendation in Alibaba"", 《ARXIV》, 24 May 2018 (2018-05-24), pages 1 - 10 * |
| 坚强: ""KDD Cup 2020 Debiasing比赛冠军技术方案及在美团的实践"", Retrieved from the Internet <URL:https://tech.meituan.com/2020/08/20/kdd-cup-debiasing-practice.html> * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Schneider et al. | Forecasting sales of new and existing products using consumer reviews: A random projections approach | |
| Wang et al. | Learning hierarchical representation model for nextbasket recommendation | |
| US11315032B2 (en) | Method and system for recommending content items to a user based on tensor factorization | |
| US11126660B1 (en) | High dimensional time series forecasting | |
| He et al. | Correlated matrix factorization for recommendation with implicit feedback | |
| CN111523010A (en) | Recommendation method and device, terminal equipment and computer storage medium | |
| CN111144933B (en) | Commodity recommendation method and device, electronic equipment and storage medium | |
| CN111737578B (en) | Recommendation method and system | |
| JP6967462B2 (en) | Information processing equipment, information processing methods, and information processing programs | |
| CN112488863B (en) | Dangerous seed recommendation method and related equipment in user cold start scene | |
| US20190080352A1 (en) | Segment Extension Based on Lookalike Selection | |
| CN111967924A (en) | Commodity recommendation method, commodity recommendation device, computer device, and medium | |
| CN110598120A (en) | Behavior data based financing recommendation method, device and equipment | |
| CN111695024A (en) | Object evaluation value prediction method and system, and recommendation method and system | |
| CN111429214B (en) | Transaction data-based buyer and seller matching method and device | |
| CN115631012A (en) | Target recommendation method and device | |
| CN107451157B (en) | Abnormal data identification method, device and system, and searching method and device | |
| CN110020918B (en) | Recommendation information generation method and system | |
| CN109886299B (en) | User portrait method and device, readable storage medium and terminal equipment | |
| CN116578774A (en) | Method, device, computer equipment and storage medium for pre-estimated sorting | |
| CN114881108A (en) | Commodity recommendation recall method, device, equipment and storage medium | |
| Cherednichenko et al. | Item Matching Model in E-Commerce: How Users Benefit | |
| Lu | Knowledge distillation-enhanced multitask framework for recommendation | |
| CN112184250B (en) | Method and device for generating retrieval page, storage medium and computer equipment | |
| Tian et al. | I2PS: An Online Item Intelligent Publishing System in Consumer to Consumer (C2C) Transaction Platform |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |