CN115099885A - Commodity matching recommendation method and system - Google Patents
Commodity matching recommendation method and system Download PDFInfo
- Publication number
- CN115099885A CN115099885A CN202210346071.1A CN202210346071A CN115099885A CN 115099885 A CN115099885 A CN 115099885A CN 202210346071 A CN202210346071 A CN 202210346071A CN 115099885 A CN115099885 A CN 115099885A
- Authority
- CN
- China
- Prior art keywords
- commodity
- customer
- fitness
- client
- matrix
- 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 45
- 239000011159 matrix material Substances 0.000 claims abstract description 124
- 241000282461 Canis lupus Species 0.000 claims description 29
- 238000003062 neural network model Methods 0.000 claims description 26
- 238000005457 optimization Methods 0.000 claims description 21
- 238000013507 mapping Methods 0.000 claims description 9
- 230000010354 integration Effects 0.000 claims description 8
- 238000005034 decoration Methods 0.000 claims description 6
- 239000013598 vector Substances 0.000 claims description 4
- 238000010411 cooking Methods 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 8
- 230000000739 chaotic effect Effects 0.000 description 4
- 230000007246 mechanism Effects 0.000 description 4
- 238000013461 design Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 230000006978 adaptation Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 239000006185 dispersion Substances 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004821 distillation Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000009191 jumping Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- 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
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
Abstract
The invention discloses a commodity matching recommendation method and a commodity matching recommendation system, which are characterized in that customer information is acquired; generating a customer preference matrix according to the customer information; calculating the Euclidean distance between each commodity feature matrix and a customer preference matrix in a commodity database; finding out a commodity feature matrix with the minimum Euclidean distance from the customer preference matrix; recommending the commodity information corresponding to the commodity feature matrix to a customer; therefore, the commodity recommending method and the commodity recommending system recommend the commodity with the highest matching degree with the customer to the customer, realize accurate commodity recommendation for the customer, and solve the technical problem that the commodity cannot be accurately recommended in the prior art.
Description
Technical Field
The invention belongs to the technical field of data, and particularly relates to a commodity matching recommendation method and system.
Background
In recent years, the high-speed development of the e-commerce enables the national express business volume to maintain the acceleration of more than 20% for a long time, the physical e-commerce develops into a new round of ascending channel, and the new kinetic energy of consumption logistics is continuously increased. Meanwhile, research shows that 27.2% of consumers express the desire to enrich the delivered services, namely, improve and upgrade the logistics services after delivery is completed. In this context, the scenario logistics operation is performed, and the logistics enterprise interacts with the user through the delivery contact, knows the consumer appeal, and provides a full scenario solution for the user together with the partner on the supply chain.
The current commodity recommendation strategy cannot accurately acquire the potential demands of customers, cannot accurately recommend commodities, and is easy to generalize an output scheme and lack of pertinence.
Disclosure of Invention
The invention provides a commodity matching recommendation method, which solves the problem that commodities cannot be accurately recommended in the prior art.
In order to solve the technical problems, the invention adopts the following technical scheme:
a commodity matching recommendation method comprises the following steps:
acquiring customer information;
generating a client preference matrix according to the client information;
calculating the Euclidean distance between each commodity feature matrix in the commodity database and the customer preference matrix; finding out a commodity feature matrix with the minimum Euclidean distance from the customer preference matrix;
and recommending the commodity information corresponding to the commodity feature matrix to the customer.
In some embodiments of the present application, the customer information includes one or more of family members, presence or absence of an infant or pregnant woman, city where the house is located, location of the house, area of the house, recent presence or absence of decoration or decoration plan, presence or absence of fixed home, frequent cooking, willingness to purchase smart homes, presence or absence of home fitness, travel mode, and price of goods willing to purchase in the same type of goods.
In some embodiments of the present application, the method further comprises:
obtaining feedback information of a client;
if the feedback information is that the customer does not purchase the recommended commodity, updating the commodity database, and recalculating the Euclidean distance between each commodity feature matrix in the commodity database and the customer preference matrix; finding out a commodity feature matrix with the minimum Euclidean distance from the customer preference matrix; and recommending the commodity information corresponding to the commodity feature matrix to the customer.
In some embodiments of the present application, the commodity feature matrix with the minimum euclidean distance to the customer preference matrix is obtained by a bubble sorting method.
In some embodiments of the present application, the generating a customer preference matrix according to customer information specifically includes:
converting the client information into client characteristic codes through one-hot coding;
and inputting the client feature code into a neural network model to obtain a client preference matrix.
In some embodiments of the present application, the neural network model comprises:
an encoder comprising a plurality of pyramid structures and an integration structure, each pyramid structure comprising a plurality of multi-headed sparse self-attention layers stacked in sequence; the input end of each pyramid structure receives the customer feature code, and the output end of each pyramid structure is connected with the input end of the integration structure;
a decoder comprising a multi-headed sparsity self-attention layer and a multi-headed attention layer; the multi-headed sparsity of the decoder receives client feature codes from an input of an attention layer; the output end of the multi-head sparsity self-attention layer of the decoder is connected with the first input end of the multi-head attention layer, and the second input end of the multi-head attention layer is connected with the output end of the integrated structure;
the input end of the full connection layer is connected with the output end of the multi-head attention layer of the decoder; and the output end of the full connection layer outputs a client preference matrix.
In some embodiments of the present application, the neural network model is a cgwformer self-optimizing neural network model, further comprising:
and the self-optimization module automatically adjusts the hyper-parameters by adopting a CGWO normal cloud wolf optimization algorithm.
In some embodiments of the present application, the normal cloud wolf optimization algorithm includes the following steps:
(1) initializing a population by adopting Tent mapping in a search space;
(2) calculating the fitness value of each wolf individual and sequencing; obtaining fitness values fitness alpha, fitness beta and fitness delta of the first three in the arrangement and corresponding three individual positions X alpha, X beta and X delta;
(3) updating the convergence factor a and the values of the coefficient vectors A and C;
(4) updating the position of the wolf group, and finding out the optimal individual position;
(5) obtaining the position X' of the wolf group at the moment through a normal cloud generator; recalculating the fitness value of each individual, and finding out the fitness values fitness 'alpha, fitness' beta and fitness 'delta of the first three in the arrangement and corresponding three individual positions X' alpha, X 'beta and X' delta;
(6) comparing the fitness ' alpha, the fitness ' beta and the fitness ' delta with the fitness alpha, the fitness beta and the fitness delta, and updating X alpha, X beta and X delta; adding 1 to the iteration times;
(7) judging whether the maximum iteration times is reached;
if the maximum iteration number is not reached, returning to the step (2);
and if the maximum iteration number is reached, outputting the X alpha as the optimal solution.
A merchandise matching recommendation system comprising:
the client information acquisition module is used for acquiring client information;
the client preference matrix generating module is used for generating a client preference matrix according to the client information;
the matching module is used for calculating the Euclidean distance between each commodity feature matrix in the commodity database and the customer preference matrix; finding out a commodity feature matrix with the minimum Euclidean distance from the customer preference matrix;
and the recommendation output module is used for recommending the commodity information corresponding to the commodity feature matrix with the minimum Euclidean distance of the customer preference matrix to the customer.
In some embodiments of the present application, the client information collection module and the recommendation output module are disposed on an intelligent terminal.
Compared with the prior art, the invention has the advantages and positive effects that: according to the commodity matching recommendation method and system, the client information is obtained; generating a client preference matrix according to the client information; calculating the Euclidean distance between each commodity feature matrix and a customer preference matrix in a commodity database; finding out a commodity feature matrix with the minimum Euclidean distance from the customer preference matrix; recommending the commodity information corresponding to the commodity feature matrix to a customer; therefore, the commodity recommending method and the commodity recommending system recommend the commodity with the highest matching degree with the customer to the customer, realize accurate commodity recommendation for the customer, and solve the technical problem that the commodity cannot be accurately recommended in the prior art.
Other features and advantages of the present invention will become more apparent from the detailed description of the embodiments of the present invention when taken in conjunction with the accompanying drawings.
Drawings
FIG. 1 is a flowchart of an embodiment of a method for recommending matching of goods according to the present invention;
fig. 2 is a flowchart of another embodiment of the method for recommending a product match according to the present invention;
FIG. 3 is a schematic diagram of the structure of one embodiment of a neural network model;
FIG. 4 is a graph of the distribution of cloud droplets in a normal cloud model as a function of numerical characteristic values;
FIG. 5 is a flow diagram of one embodiment of a gray wolf optimization algorithm;
FIG. 6 is a flowchart of another embodiment of a method for recommending matching of goods according to the present invention;
fig. 7 is a block diagram of an embodiment of a product matching recommendation system according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the accompanying drawings and examples.
It should be noted that in the description of the present invention, the terms of direction or positional relationship indicated by the terms "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. are based on the directions or positional relationships shown in the drawings, which are merely for convenience of description, and do not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation, and thus, should not be construed as limiting the present invention. Furthermore, the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
Furthermore, it should be noted that, in the description of the present invention, unless otherwise explicitly specified or limited, the terms "mounted," "connected," and "connected" are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood by those skilled in the art according to specific situations.
Aiming at the technical problem that the commodities cannot be accurately recommended to the client at present, the invention provides a commodity matching recommendation method and system, and the commodities are accurately recommended to the client. Hereinafter, the commodity matching recommendation method and system of the present invention will be described in detail with reference to the accompanying drawings.
The first embodiment,
The commodity matching recommendation method of the embodiment mainly includes the following steps, which are shown in fig. 1.
Step S11: and acquiring the client information.
In some embodiments of the present application, the customer information includes one or more of family members, presence or absence of an infant or a pregnant woman, a city where the house is located, a section where the house is located, an area of the house, presence or absence of recent decoration or decoration plan, presence or absence of fixed home, frequent cooking, presence or absence of willingness to purchase smart homes, presence or absence of home fitness, a travel mode, a price of a commodity willing to purchase in a same kind of commodity, and the like.
Through the information, the user portrait can be accurately constructed, the client can be comprehensively known, the requirement of the client is known, and the accuracy of commodity recommendation can be further improved.
Step S12: and generating a customer preference matrix according to the customer information.
In some embodiments of the present application, the step specifically includes the following steps:
and (12-1) converting the client information into the client characteristic code through one-hot coding.
And converting the client information into the one-hot code of (22, 1) according to a preset dimension conversion module to generate the client feature code.
And (12-2) inputting the client feature codes into the neural network model to obtain a client preference matrix.
And converting the client characteristic code into a client preference matrix unique to the client through a trained neural network model. The client preference matrix is in a one-hot encoding format.
The neural network model has self-learning performance, and the trained neural network model can convert the client feature codes into the unique client preference matrix meeting the requirements, so that the foundation is laid for subsequent accurate recommendation.
Recommending a client adaptation product based on the neural network technology, and generating a client preference matrix according to client information to represent the adaptation product characteristics of the client.
The initial learning rate is 0.0001, 20 epochs, and the neural network model is trained and stored.
Step S13: calculating the Euclidean distance between each commodity feature matrix and a customer preference matrix in a commodity database; and finding the commodity feature matrix with the minimum Euclidean distance from the customer preference matrix.
The commodity database stores a plurality of commodity feature matrixes, and each commodity feature matrix corresponds to a commodity. Traversing the commodity database, and calculating the Euclidean distance between each commodity feature matrix in the commodity database and the customer preference matrix; the smaller the Euclidean distance is, the higher the matching degree of the commodity feature matrix and the customer preference matrix is. Therefore, the commodity feature matrix with the minimum Euclidean distance from the customer preference matrix is the commodity feature matrix with the highest matching degree with the customer preference matrix.
The calculation formula of the Euclidean distance is as follows:
wherein d is Euclidean distance, k is an integer from 1 to n, and n represents a feature matrix dimension;
x 1 representing a commodity feature matrix;
x 1k a k-th element representing a commodity feature matrix;
x 2 representing a customer preference matrix;
x 2k the kth element of the customer preference matrix is represented.
The matching degree of the commodity characteristic matrix and the customer preference matrix is judged through the Euclidean distance, so that the judgment is accurate, the calculation is simple, and the realization is convenient.
In some embodiments of the present application, the commodity feature matrix with the minimum euclidean distance to the customer preference matrix is obtained by a bubble sorting method.
The bubble sorting method is simple, stable and reliable, and can accurately find out the commodity feature matrix with the minimum Euclidean distance from the customer preference matrix.
Step S14: and recommending the commodity information corresponding to the commodity feature matrix to the customer.
Recommending the commodity information corresponding to the commodity feature matrix with the minimum Euclidean distance with the customer preference matrix to the customer.
Recommended forms include, but are not limited to, the following: the service soldier directly recommends to the client, sends the recommendation result to the client through the public number, recommends the commodity to the client through the dm advertisement, recommends to the client with the short message, etc.
According to the commodity matching recommendation method, the client information is obtained; generating a client preference matrix according to the client information; calculating the Euclidean distance between each commodity feature matrix in the commodity database and the customer preference matrix; finding out a commodity feature matrix with the minimum Euclidean distance from the customer preference matrix; recommending the commodity information corresponding to the commodity feature matrix to a customer; therefore, the commodity recommendation method of the embodiment recommends the commodity which is the highest in matching degree with the customer to the customer, so that the commodity is accurately recommended to the customer, and the technical problem that the commodity cannot be accurately recommended in the prior art is solved.
After the commodity is recommended to the client, the feedback of the client needs to be followed in time.
In some embodiments of the present application, the method for recommending matching of goods further includes the following steps, which are shown in fig. 2.
Step S15: feedback information of the client is obtained.
The feedback information includes presence or absence of purchase of recommended goods, advice, and the like.
After the customer obtains the recommendation of the related goods, if the pushing result meets the customer requirement, the customer can purchase the recommended goods online or offline, and the platform records the customer information and the purchasing result. The feedback information of the customer can be obtained through the commodity sales platform.
If the purchase record exists, the feedback information shows that the recommended commodity is purchased by the customer, and the recommended commodity meets the user requirement, namely the recommendation result is effective.
If the purchase record does not exist, the feedback information indicates that the customer does not purchase the recommended commodity, the customer does not have corresponding feedback on the recommended commodity, the recommended commodity does not meet the customer requirement, and the recommendation result is invalid.
Step S16: if the feedback information of the customer is that the customer does not purchase the recommended commodity, updating the commodity database, and then recalculating the Euclidean distance between each commodity feature matrix in the commodity database and the customer preference matrix; finding out a commodity feature matrix with the minimum Euclidean distance from the customer preference matrix; and recommending the commodity information corresponding to the commodity feature matrix to the customer.
Therefore, when the feedback information indicates that the customer has not purchased the recommended product, the product database is updated, and the process returns to step S13.
The commodity database and the neural network model can be optimized according to the feedback information of the client. When the feedback of the client is not good, the client opinion can be inquired, the commodity structure can be optimized, and the commodity database can be updated. After the customer has good feedback and purchases the goods, the sample database can be updated for the neural network model self-learning, as shown in fig. 6.
Through the design steps S15-S16, the customer feedback is timely known, the commodity database is updated according to the customer feedback, namely the commodity database is continuously enriched by using the demand information provided by the customer, and the positioning and classification of the commodities are further accurately updated, so that the experience iteration of the customer is realized, the iteration of commodity recommendation is realized, and diversified and variable demands are met.
The traditional Transformer neural network has the potential of improving the prediction capability. However, the Transformer neural network has some serious problems, such as: a secondary time complexity; high memory usage; inherent limitations of the encoder-decoder architecture; predicting the accuracy dip of the long output: the dynamic decoding process results in a very slow learning step by step and a severe accumulation of errors.
The cgwformer neural network can effectively solve the above problems, and has three significant characteristics:
(1) the sparsity self-attention mechanism achieves O (LlogL) in time complexity and memory utilization rate for an input sequence with the length of L, and has higher performance in dependence alignment of the sequence. And (3) by using the assumption that the dot product result obeys long-tail distribution, when calculating the sparsity score of each sequence, calculating only partial key values which are sampled.
The specific process is as follows:
(11) the default value is 5InL for the key values of the randomly sampled portions per sequence.
(12) A sparsity score is calculated for each sequence.
(13) The N sequences with the highest sparsity scores are selected.
(14) And only calculating dot product results of the N sequences and all the key values to further obtain an attention result.
(15) And (4) directly taking the average value of the input from the attention layer as the output by neglecting the rest L-N key values, thereby ensuring that the length of the input sequence and the output sequence of each sparsity self-attention layer is L.
(2) Data distillation based on the self-attention mechanism: control attention is highlighted by halving the encoder input through convolution and pooling, and very long input sequences are efficiently handled.
(3) The self-generating encoder is used for predicting the long-time sequence, and the structure of the sequence target sequence is realized by using the mask instead of being carried out step by step, so that the reasoning speed of the long-time sequence prediction is greatly improved.
In some embodiments of the present application, the neural network model includes an encoder, a decoder, and a fully-connected layer, as shown in fig. 3.
An encoder comprising a plurality of pyramid structures and an integration structure, each pyramid structure comprising a plurality of multi-headed sparse self-attention layers stacked in sequence; the input end of each pyramid structure receives the customer feature code, and the output end of each pyramid structure is connected with the input end of the integration structure. The integration structure is used to integrate the outputs of the plurality of pyramid structures together and then output to the decoder.
A decoder comprising a multi-headed sparsity self-attention layer and a multi-headed attention layer; the multi-head sparsity of the decoder receives a client feature code from an input end of an attention layer; the output end of the multi-head sparsity self-attention layer of the decoder is connected with the first input end of the multi-head attention layer, and the second input end of the multi-head attention layer is connected with the output end of the integration structure.
The input end of the full connection layer is connected with the output end of the multi-head attention layer of the decoder; and the output end of the full connection layer outputs a client preference matrix.
The neural network model with the structure is selected, so that the client preference matrix can be accurately and quickly generated.
In some embodiments of the present application, the neural network model is a cgwformer self-optimizing neural network model, further comprising: and the self-optimization module automatically adjusts the hyper-parameters by adopting a CGWO normal cloud wolf optimization algorithm.
That is, the neural network model is a cgwformer self-optimizing neural network model, which includes an encoder, a decoder, a full link layer, and a self-optimizing module.
The self-optimization module automatically adjusts the hyper-parameters by adopting a CGWO normal cloud wolf optimization algorithm, namely population initialization is carried out by utilizing Tent mapping in the traditional wolf optimization algorithm, and the wolf population position is determined by utilizing a normal cloud model.
The Cgwormer self-optimization neural network model with the structure is selected, so that the client preference matrix can be accurately and quickly generated.
In some embodiments of the present application, the neural network model adopts a Grey Wolf optimization algorithm (GWO), population initialization is performed in the Grey Wolf optimization algorithm by using Tent mapping, and a Wolf group position is obtained by using a normal cloud model.
Because Bayesian optimization is difficult to process high-dimensional feature vectors, and more structures of the Cgwormer neural network are more complex compared with a traditional neural network module, but the time complexity is reduced, a large amount of training can be performed to collect a hyper-parameter set and corresponding residual error results, and therefore a cloud model gray wolf optimization algorithm is adopted to solve the multi-dimensional black box optimization problem.
The cloud model can realize uncertain conversion between quantitative values and qualitative concepts, and the model can well describe and process the fuzziness and randomness of data. The normal cloud model is a cloud model which is most consistent with the random probability distribution in nature and has important mathematical significance.
The cloud model is characterized by three mathematical parameters of expectation Ex, entropy En and super-entropy He. The relationship between the distribution of cloud droplets in the normal cloud model and the numerical characteristic value is shown in fig. 4. It can be seen that the cloud drop distribution range is enlarged with the increase of En, and the dispersion degree is enlarged with the increase of He, so that the randomness and the fuzziness of the cloud drop distribution are well reflected. The normal forward cloud generator is an algorithm that produces cloud droplets that are substantially normally distributed, one cloud droplet per run, until the desired number of cloud droplets are produced.
The more uniform the distribution of the initial population in the search space, the more beneficial the optimization efficiency and the solving precision of the algorithm are. The chaotic sequence has the characteristics of good randomness, ergodicity and regularity, and the basic principle is that the chaotic sequence is generated between [0 and 1] through a mapping relation and then is converted into an individual search space. Tent mapping can generate a more evenly distributed sequence than other mappings, and is therefore used to initialize the wolf population. Tent map is mathematically represented as follows:
wherein: mu is a chaotic parameter which is in direct proportion to the chaos; i and j are the population number and the chaotic variable sequence number respectively; y is the corresponding value.
In order to further improve GWO the ability of jumping out of the local optimal solution when processing the complex optimization problem and improve the convergence precision of the algorithm, a normal cloud model is introduced as a new updating mechanism of the wolf pack position on the basis of population initialization by Tent mapping, in the new mechanism, the current optimal individual is selected in the prey attack stage, and the position of the optimal individual is taken as the expected value Ex of the normal cloud model for deep development, and the new position updating formula is as follows:
Position’＝Gnc(position_best,En,He,Nd)；
wherein, the position _ best is the position of the current optimal individual. Nd is the number of cloud drops that the desired cloud model generates. Therefore, the range of the updated position of the wolf colony from the optimal individual can be adjusted through the value of En, and the dispersion degree of updating the position of the wolf colony is controlled through He. Generally, in the early stage of a wolf pack hunting, the wolf pack is far from the food location, and a larger location update range can be selected at this time. With the increase of the iteration times, the wolf pack approaches to food gradually, so that the updating range can be gradually reduced, and the searching precision is improved. According to the wolf group predation process, the value of En can be adaptively adjusted as follows:
wherein, omega is a real number between [0,1], maximer is the maximum iteration number, T is the current iteration number, and T is a positive integer.
In some embodiments of the present application, the normal cloud wolf optimization algorithm includes the following steps, as shown in fig. 5.
First, algorithm parameters are initialized.
Step S21: and initializing the population by Tent mapping in the search space.
Step S22: calculating the fitness value of each wolf individual and sequencing; fitness values fitness α, fitness β, and fitness δ of the first three of the permutations and the corresponding three individual positions X α, X β, and X δ are obtained.
Step S23: the convergence factor a of the algorithm and the values of the coefficient vectors a and C are updated GWO.
Step S24: updating the position of the wolf pack based on a standard GWO algorithm, and finding out the optimal individual position, namely position _ best.
Step S25: obtaining the position X' of the wolf group at the moment through a normal cloud generator; and recalculating the fitness value of each individual, and finding out the fitness values fitness 'alpha, fitness' beta and fitness 'delta of the first three in the arrangement and the corresponding three individual positions X' alpha, X 'beta and X' delta.
Step S26: comparing the fitness ' alpha, the fitness ' beta and the fitness ' delta with the fitness alpha, the fitness beta and the fitness delta, and updating X alpha, X beta and X delta; the number of iterations is increased by 1.
Step S27: and judging whether the maximum iteration number is reached.
If the current iteration number does not reach the maximum iteration number, the process returns to step S22.
If the current iteration number reaches the maximum iteration number, the algorithm is ended, and step S28 is executed: and outputting the X alpha as an optimal solution.
And outputting the optimal solution for updating the hyperparameters of the Cgwormer neural network (for improving the training effect of the neural network).
Through the design steps S21-S28, the optimal solution is quickly found out so as to optimize the neural network model.
According to the recommendation matching method, the Cgwoformer neural network model is adopted to generate an accurate client preference matrix so as to further improve the accuracy of commodity recommendation.
Respectively training and evaluating six models including Cgformer based on sample data obtained by issuing questionnaires, wherein the evaluation standard is the average residual error of a client preference matrix, and the final result is as follows:
the lower the MSE, the more accurate the prediction of client preference.
Example II,
Based on the design of the commodity matching recommendation method in the first embodiment, the second embodiment further provides a commodity matching recommendation system, which includes a customer information acquisition module, a customer preference matrix generation module, a matching module, and a recommendation output module, as shown in fig. 7.
And the client information acquisition module is used for acquiring the client information.
And the client preference matrix generating module is used for generating a client preference matrix according to the client information.
The matching module is used for calculating the Euclidean distance between each commodity feature matrix in the commodity database and the customer preference matrix; and finding out the commodity feature matrix with the minimum Euclidean distance to the customer preference matrix.
And the recommendation output module is used for recommending the commodity information corresponding to the commodity feature matrix with the minimum Euclidean distance with the customer preference matrix to the customer.
In some embodiments of the application, the client information acquisition module and the recommendation output module are arranged on an intelligent terminal, such as a smart phone, a PAD, and the like, so that the client information acquisition module and the recommendation output module are convenient to acquire client information and recommend commodities to clients. And is presented to the customer in the form of an APP.
The working process of the specific product matching recommendation system has been described in detail in the product matching recommendation method in the first embodiment, and is not described herein again.
The commodity matching recommendation system of the embodiment acquires the customer information; generating a customer preference matrix according to the customer information; calculating the Euclidean distance between each commodity feature matrix in the commodity database and the customer preference matrix; finding out a commodity feature matrix with the minimum Euclidean distance from the customer preference matrix; recommending the commodity information corresponding to the commodity feature matrix to a customer; therefore, the commodity recommendation system of the embodiment recommends the commodity with the highest matching degree with the customer to the customer, so as to realize accurate commodity recommendation for the customer, and solve the technical problem that the commodity cannot be accurately recommended in the prior art.
It is to be understood that the above description is not intended to limit the present invention, and the present invention is not limited to the above examples, and those skilled in the art may make modifications, alterations, additions or substitutions within the spirit and scope of the present invention.
Claims (10)
1. A commodity matching recommendation method is characterized in that: the method comprises the following steps:
acquiring customer information;
generating a client preference matrix according to the client information;
calculating the Euclidean distance between each commodity feature matrix in the commodity database and the customer preference matrix; finding out a commodity feature matrix with the minimum Euclidean distance from the customer preference matrix;
and recommending the commodity information corresponding to the commodity feature matrix to the customer.
2. The commodity matching recommendation method according to claim 1, characterized in that: the client information comprises one or more of family members, existence of infants or pregnant women, cities where houses are located, places where houses are located, house areas, recent existence of decorations or decoration plans, existence of fixed housekeeping, frequent cooking, existence of willingness to purchase intelligent homes, existence of family fitness, travel modes and prices of commodities willing to be purchased in similar commodities.
3. The merchandise matching recommendation method according to claim 1, characterized in that: the method further comprises the following steps:
obtaining feedback information of a client;
if the feedback information is that the customer does not purchase the recommended commodity, updating the commodity database, and recalculating the Euclidean distance between each commodity feature matrix in the commodity database and the customer preference matrix; finding out a commodity feature matrix with the minimum Euclidean distance from the customer preference matrix; and recommending the commodity information corresponding to the commodity feature matrix to the customer.
4. The commodity matching recommendation method according to claim 1, characterized in that: and obtaining a commodity feature matrix with the minimum Euclidean distance from the customer preference matrix by a bubble sorting method.
5. The merchandise matching recommendation method according to claim 1, characterized in that: the generating a customer preference matrix according to the customer information specifically includes:
converting the client information into client characteristic codes through one-hot coding;
and inputting the client feature code into a neural network model to obtain a client preference matrix.
6. The commodity matching recommendation method according to claim 5, characterized in that: the neural network model includes:
an encoder comprising a plurality of pyramid structures and an integration structure, each pyramid structure comprising a plurality of multi-headed sparse self-attention layers stacked in sequence; the input end of each pyramid structure receives the customer feature code, and the output end of each pyramid structure is connected with the input end of the integration structure;
a decoder comprising a multi-headed sparsity self-attention layer and a multi-headed attention layer; the multi-headed sparsity of the decoder receives client feature codes from an input of an attention layer; the output end of the multi-head sparsity self-attention layer of the decoder is connected with the first input end of the multi-head attention layer, and the second input end of the multi-head attention layer is connected with the output end of the integrated structure;
a full connection layer, the input end of which is connected with the output end of the multi-head attention layer of the decoder; and the output end of the full connection layer outputs a client preference matrix.
7. The commodity matching recommendation method according to claim 6, characterized in that: the neural network model is a Cgwormer self-optimizing neural network model, and further comprises:
and the self-optimization module automatically adjusts the hyper-parameters by adopting a CGWO normal cloud wolf optimization algorithm.
8. The commodity matching recommendation method according to claim 7, characterized in that: the normal cloud wolf optimization algorithm comprises the following steps:
(1) initializing a population by adopting Tent mapping in a search space;
(2) calculating the fitness value of each wolf individual and sequencing; obtaining fitness values fitness alpha, fitness beta and fitness delta of the first three in the arrangement and corresponding three individual positions X alpha, X beta and X delta;
(3) updating the convergence factor a and the values of the coefficient vectors A and C;
(4) updating the position of the wolf pack to find out the optimal individual position;
(5) obtaining the position X' of the wolf gray colony at the moment through a normal cloud generator; recalculating the fitness value of each individual, and finding out the fitness values fitness 'alpha, fitness' beta and fitness 'delta of the first three in the arrangement and corresponding three individual positions X' alpha, X 'beta and X' delta;
(6) comparing the fitness ' alpha, the fitness ' beta and the fitness ' delta with the fitness alpha, the fitness beta and the fitness delta, and updating X alpha, X beta and X delta; adding 1 to the iteration times;
(7) judging whether the maximum iteration times is reached;
if the maximum iteration number is not reached, returning to the step (2);
and if the maximum iteration number is reached, outputting the X alpha as the optimal solution.
9. A commodity matching recommendation system is characterized in that: the method comprises the following steps:
the client information acquisition module is used for acquiring client information;
the client preference matrix generating module is used for generating a client preference matrix according to the client information;
the matching module is used for calculating the Euclidean distance between each commodity feature matrix in the commodity database and the customer preference matrix; finding out a commodity feature matrix with the minimum Euclidean distance from the customer preference matrix;
and the recommendation output module is used for recommending the commodity information corresponding to the commodity feature matrix with the minimum Euclidean distance with the customer preference matrix to the customer.
10. The merchandise matching recommendation system according to claim 9, characterized in that: the client information acquisition module and the recommendation output module are arranged on the intelligent terminal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210346071.1A CN115099885A (en) | 2022-03-31 | 2022-03-31 | Commodity matching recommendation method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210346071.1A CN115099885A (en) | 2022-03-31 | 2022-03-31 | Commodity matching recommendation method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115099885A true CN115099885A (en) | 2022-09-23 |
Family
ID=83286995
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210346071.1A Pending CN115099885A (en) | 2022-03-31 | 2022-03-31 | Commodity matching recommendation method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115099885A (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110503498A (en) * | 2018-05-16 | 2019-11-26 | 北京三快在线科技有限公司 | A kind of order recommended method and device |
| CN111028029A (en) * | 2018-10-10 | 2020-04-17 | 深圳云天励飞技术有限公司 | Offline commodity recommendation method and device and electronic equipment |
| CN112653142A (en) * | 2020-12-18 | 2021-04-13 | 武汉大学 | Wind power prediction method and system for optimizing depth transform network |
| WO2021128510A1 (en) * | 2019-12-27 | 2021-07-01 | 江苏科技大学 | Bearing defect identification method based on sdae and improved gwo-svm |
| CN113241122A (en) * | 2021-06-11 | 2021-08-10 | 长春工业大学 | Gene data variable selection and classification method based on fusion of adaptive elastic network and deep neural network |
| CN113449680A (en) * | 2021-07-15 | 2021-09-28 | 北京理工大学 | Knowledge distillation-based multimode small target detection method |
| CN114169927A (en) * | 2021-12-07 | 2022-03-11 | 合肥工业大学 | Product personalized combination recommendation method based on multi-arm slot machine algorithm |
-
2022
- 2022-03-31 CN CN202210346071.1A patent/CN115099885A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110503498A (en) * | 2018-05-16 | 2019-11-26 | 北京三快在线科技有限公司 | A kind of order recommended method and device |
| CN111028029A (en) * | 2018-10-10 | 2020-04-17 | 深圳云天励飞技术有限公司 | Offline commodity recommendation method and device and electronic equipment |
| WO2021128510A1 (en) * | 2019-12-27 | 2021-07-01 | 江苏科技大学 | Bearing defect identification method based on sdae and improved gwo-svm |
| CN112653142A (en) * | 2020-12-18 | 2021-04-13 | 武汉大学 | Wind power prediction method and system for optimizing depth transform network |
| CN113241122A (en) * | 2021-06-11 | 2021-08-10 | 长春工业大学 | Gene data variable selection and classification method based on fusion of adaptive elastic network and deep neural network |
| CN113449680A (en) * | 2021-07-15 | 2021-09-28 | 北京理工大学 | Knowledge distillation-based multimode small target detection method |
| CN114169927A (en) * | 2021-12-07 | 2022-03-11 | 合肥工业大学 | Product personalized combination recommendation method based on multi-arm slot machine algorithm |
Non-Patent Citations (1)
| Title |
|---|
| 张铸 等: "基于自适应正态云模型的灰狼优化算法", 控制与决策, 3 July 2020 (2020-07-03), pages 2563 - 2564 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Guan et al. | Matrix factorization with rating completion: An enhanced SVD model for collaborative filtering recommender systems | |
| Kurumatani | Time series forecasting of agricultural product prices based on recurrent neural networks and its evaluation method | |
| CN101783004A (en) | Fast intelligent commodity recommendation system | |
| Vendrov et al. | Gradient-based optimization for bayesian preference elicitation | |
| CN111191814A (en) | Electricity price prediction method, system and computer readable storage medium | |
| CN114564651A (en) | Self-supervision recommendation method combined with contrast learning method | |
| CN109146553A (en) | Spot Price forecasting system and its method based on multi-density cluster and multicore SVM | |
| CN115374288A (en) | Recommendation method based on knowledge graph and attention mechanism | |
| Chen et al. | An ensemble feature selection method for short-term electrical load forecasting | |
| Huynh et al. | Multi-output Gaussian processes for multi-population longevity modelling | |
| Afshar et al. | A decision support method to increase the revenue of ad publishers in waterfall strategy | |
| CN106991496A (en) | A kind of user behavior towards mobile social environment is layered interaction prediction method | |
| CN112182498B (en) | Old people nursing device and method based on network representation learning | |
| CN115099885A (en) | Commodity matching recommendation method and system | |
| CN116934531A (en) | Wine information intelligent management method and system based on data analysis | |
| CN111506813A (en) | Remote sensing information accurate recommendation method based on user portrait | |
| CN116304374A (en) | Customer matching method and system based on package data | |
| CN110070099A (en) | A kind of industrial data feature structure method based on intensified learning | |
| CN116362836A (en) | Agricultural product recommendation algorithm based on user behavior sequence | |
| Ilyas et al. | Comparative Analysis of Regression Algosrithms used to Predict the Sales of Big Marts | |
| Khakpour | Data science for decision support: Using machine learning and big data in sales forecasting for production and retail | |
| Chang et al. | An interval-valued time series forecasting scheme with probability distribution features for electric power generation prediction | |
| Yan et al. | Modeling long-and short-term user behaviors for sequential recommendation with deep neural networks | |
| Ibrahim et al. | Short-term multivariate time series load data forecasting at low-voltage level using optimised deep-ensemble learning-based models | |
| CN106021590B (en) | B2B platform supplier recommendation method and system |
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 |