CN117635238A - Commodity recommendation method, device, equipment and storage medium - Google Patents
Commodity recommendation method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN117635238A CN117635238A CN202311782104.8A CN202311782104A CN117635238A CN 117635238 A CN117635238 A CN 117635238A CN 202311782104 A CN202311782104 A CN 202311782104A CN 117635238 A CN117635238 A CN 117635238A
- Authority
- CN
- China
- Prior art keywords
- enterprise
- commodity
- model
- similarity
- data
- 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 58
- 239000013598 vector Substances 0.000 claims abstract description 127
- 230000007246 mechanism Effects 0.000 claims description 30
- 238000012549 training Methods 0.000 claims description 22
- 238000004590 computer program Methods 0.000 claims description 14
- 230000006870 function Effects 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 10
- 230000006399 behavior Effects 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 7
- 230000003068 static effect Effects 0.000 claims description 7
- 238000010586 diagram Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 8
- 238000012163 sequencing technique Methods 0.000 description 8
- 238000010276 construction Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 7
- 238000004891 communication Methods 0.000 description 6
- 230000009471 action Effects 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000013526 transfer learning Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 210000004899 c-terminal region Anatomy 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000005714 functional activity Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
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
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0255—Targeted advertisements based on user history
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06N3/08—Learning methods
-
- 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
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0269—Targeted advertisements based on user profile or attribute
- G06Q30/0271—Personalized advertisement
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Accounting & Taxation (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Entrepreneurship & Innovation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Health & Medical Sciences (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 method, a commodity recommendation device, commodity recommendation equipment and a commodity recommendation storage medium, relates to the technical field of computers, and comprises the following steps: acquiring commodity retrieval data input by a target enterprise; coding commodity retrieval data by using a coding model created based on a DSSM double-tower model to obtain retrieval vectors; calculating the similarity of each code vector in the search vector and the code vector set to obtain a similarity value; sorting the similarity values, and acquiring a preset number of similarity values from the sorted similarity values to obtain an enterprise set similar to the target enterprise; counting the purchase frequency of commodities purchased by each enterprise in the enterprise set to obtain a frequency counting result, sorting the frequency counting results, acquiring a preset number of frequency counting results from the sorted purchase frequency, and recommending commodities corresponding to the preset number of frequency counting results to a target enterprise. The commodity recommendation accuracy can be improved, the manual experience is not relied on, and the commodity recommendation method and device are suitable for scenes with fewer commodity features.
Description
Technical Field
The present disclosure relates to the field of computer technologies, and in particular, to a method, an apparatus, a device, and a storage medium for recommending commodities.
Background
At present, when recommending digital service commodities, depending on the knowledge of service personnel on certain service categories, the service personnel analyze which commodities are recommended to which users through a plurality of single and fewer factors, and the method has great limitation, and the recommendation result is often very inaccurate, so that the application effect of an actual scene is affected. For example, the method is often poor in recommending effect depending on service rules, the service rules are gradually disabled along with time and scene changes, in addition, the service depends on experience of service personnel, and the experience changes along with time and space, so that the method cannot be well applied to the current digital application scene.
While the recommendation method based on the large-scale recommendation algorithm is more in line with the current digital scene, the recommendation method is more dependent on data volume and data quality, simultaneously requires rich commodity characteristics, can be better applied to C-terminal users only by meeting the conditions, and can cause poorer recommendation effect for B-terminal users with less data volume and less behaviors. For example, in the cold start stage, a model of a similar user is obtained first, and then a user product is recommended according to the similar user, however, the similar user can build the similar model in an unsupervised clustering mode under the condition of no label, and can build the similar model through other related data transfer learning, and the final recommendation effect is also not ideal because the unsupervised learning effect is generally common but the target samples of the transfer learning are different; in addition, when the double-tower model is built, the traditional double-tower model depends on the learning capability of DNN (Deep Neural Networks, deep neural network), the DNN has good fitting property, and the phenomenon of overfitting is easy to generate on trained data, so that the prediction effect is poor, and meanwhile, the real recommendation scene has a certain feature importance sequence on different features of a user, so that the weight in the manual definition model is limited, and the recommendation effect is influenced.
In summary, how to recommend digital service type commodities is a problem that is still further solved in the field.
Disclosure of Invention
In view of the foregoing, an object of the present application is to provide a commodity recommendation method, apparatus, device and storage medium, which can improve accuracy of commodity recommendation, and is applicable to a scene with fewer commodity features without relying on manual experience. The specific scheme is as follows:
in a first aspect, the present application discloses a commodity recommendation method, including:
acquiring commodity retrieval data input by a target enterprise;
coding the commodity retrieval data by using the trained coding model to obtain a retrieval vector; the coding model is a model obtained by training a DSSM double-tower model by utilizing historical enterprise data and commodity purchase labels; the DSSM double-tower model comprises a first tower model and a second tower model, each tower model comprises a dynamic attention mechanism layer and a plurality of fully-connected layers, and the dynamic attention mechanism layer adopts a SENet structure;
respectively calculating the similarity of the retrieval vector and each code vector in a code vector set obtained by coding the historical enterprise data by the code model in advance to obtain a similarity value;
Sorting the similarity values to obtain sorted similarity, and obtaining a preset number of similarity values from the sorted similarity to obtain an enterprise set similar to the target enterprise;
counting the purchase frequency of commodities purchased by each enterprise in the enterprise set to obtain a frequency counting result, and sorting the frequency counting result to obtain a sorted purchase frequency;
and acquiring the preset number of frequency statistics results from the ordered purchase frequency, and recommending commodities corresponding to the preset number of frequency statistics results to the target enterprise.
Optionally, the calculating the similarity between the search vector and each encoding vector in the encoding vector set obtained by encoding the historical enterprise data by using the encoding model in advance, to obtain a similarity value includes:
and inputting the search vector and a code vector set obtained by coding the historical enterprise data by the code model in advance into a Hash similar search model to calculate the similarity of each code vector in the search vector and the code vector set by using a cosine distance respectively so as to obtain a similarity value.
Optionally, before the acquiring the commodity retrieval data input by the target enterprise, the method further includes:
collecting static data and dynamic behavior data of different enterprises to obtain historical enterprise data;
constructing a feature set related to a user recommendation scene based on the historical enterprise data to form a user portrait feature library;
acquiring historical commodity purchase records of different enterprises, and constructing positive and negative sample labels based on the historical commodity purchase records to obtain commodity purchase labels;
inputting the commodity purchase label and the data in the user portrait feature library into a DSSM double-tower model for training for multiple times to obtain a coding model;
and coding each characteristic data in the user portrait characteristic library by using the coding model to obtain a coding vector set.
Optionally, the constructing a feature set related to the user recommended scenario based on the historical enterprise data to form a user portrait feature library includes:
processing the historical enterprise data to obtain processed enterprise data, and extracting characteristics of the processed enterprise data to obtain enterprise characteristics;
and constructing a feature set related to the user recommendation scene based on the enterprise features to form a user portrait feature library.
Optionally, the constructing positive and negative sample labels based on the historical commodity purchase record to obtain commodity purchase labels includes:
and constructing positive and negative sample labels based on the historical commodity purchase records in a mode of the longest public substring to obtain commodity purchase labels.
Optionally, the commodity recommendation method further includes:
and optimizing model parameters of the DSSM double-tower model by using a binary cross entropy loss function.
Optionally, recommending the commodity corresponding to the preset number of frequency statistics to the target enterprise includes:
respectively determining commodities corresponding to the preset number of frequency statistics results to obtain commodities to be recommended;
and creating a commodity recommendation list based on all the commodities to be recommended, and recommending the commodity recommendation list to the target enterprise.
In a second aspect, the present application discloses a commodity recommendation device, comprising:
the first acquisition module is used for acquiring commodity retrieval data input by a target enterprise;
the coding module is used for coding the commodity retrieval data by utilizing the trained coding model to obtain a retrieval vector; the coding model is a model obtained by training a DSSM double-tower model by utilizing historical enterprise data and commodity purchase labels; the DSSM double-tower model comprises a first tower model and a second tower model, each tower model comprises a dynamic attention mechanism layer and a plurality of fully-connected layers, and the dynamic attention mechanism layer adopts a SENet structure;
The similarity calculation module is used for calculating the similarity of each code vector in the code vector set obtained by respectively searching the vector and coding the historical enterprise data by the code model in advance to obtain a similarity value;
the first ordering module is used for ordering the similarity values to obtain ordered similarity;
the second acquisition module is used for acquiring a preset number of similarity values from the ordered similarity to obtain an enterprise set similar to the target enterprise;
the statistics module is used for counting the purchase frequency of the commodities purchased by each enterprise in the enterprise set to obtain a frequency statistics result;
the second ordering module is used for ordering the frequency statistics results to obtain ordered purchasing frequency;
the third acquisition module is used for acquiring the preset number of frequency statistics results from the ordered purchase frequencies;
and the commodity recommending module is used for recommending commodities corresponding to the preset number of frequency statistics to the target enterprise.
In a third aspect, the present application discloses an electronic device comprising a processor and a memory; the processor implements the commodity recommendation method when executing the computer program stored in the memory.
In a fourth aspect, the present application discloses a computer-readable storage medium for storing a computer program; wherein the computer program when executed by the processor implements the aforementioned merchandise recommendation method.
Therefore, commodity retrieval data input by a target enterprise are firstly obtained, and then the trained coding model is utilized to code the commodity retrieval data to obtain a retrieval vector; the coding model is a model obtained by training a DSSM double-tower model by utilizing historical enterprise data and commodity purchase labels; the DSSM double-tower model comprises a first tower model and a second tower model, each tower model comprises a dynamic attention mechanism layer and a plurality of fully-connected layers, and the dynamic attention mechanism layer adopts a SENet structure; and then, respectively calculating the search vectors and the similarity of each code vector in a code vector set obtained by coding the historical enterprise data by utilizing the coding model in advance to obtain a similarity value, sequencing the similarity values to obtain sequenced similarity, acquiring a preset number of similarity values from the sequenced similarity to obtain an enterprise set similar to the target enterprise, finally counting the purchase frequency of commodities purchased by each enterprise in the enterprise set to obtain a frequency counting result, sequencing the frequency counting result to obtain sequenced purchase frequency, acquiring the preset number of frequency counting results from the sequenced purchase frequency, and recommending the commodities corresponding to the preset number of frequency counting results to the target enterprise. According to the method, the code model created based on the DSSM double-tower model is utilized to code commodity retrieval data input by a target enterprise to obtain the retrieval vector, and the similarity of the retrieval vector and each code vector in the code vector set is calculated, so that an enterprise set similar to the target enterprise is obtained, and the commodity to be recommended is determined according to the purchase frequency of the commodity purchased by each enterprise in the enterprise set, so that the commodity recommendation accuracy can be improved, the method is independent of manual experience, and the method is suitable for scenes with fewer commodity features.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present application, and that other drawings may be obtained according to the provided drawings without inventive effort to a person skilled in the art.
FIG. 1 is a flow chart of a method for recommending commodities disclosed in the present application;
FIG. 2 is a block flow diagram of a specific product recommendation method disclosed in the present application;
FIG. 3 is a schematic diagram of a particular longest common substring disclosed herein;
FIG. 4 is a schematic diagram of a DSSM double-tower model structure as disclosed herein;
FIG. 5 is a schematic diagram of a specific SENet structure disclosed in the present application;
FIG. 6 is a flowchart of a specific product recommendation method disclosed in the present application;
fig. 7 is a schematic structural diagram of a commodity recommendation device disclosed in the present application;
fig. 8 is a block diagram of an electronic device disclosed in the present application.
Detailed Description
The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all, of the embodiments of the present application. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.
The embodiment of the application discloses a commodity recommendation method, which is shown in fig. 1 and comprises the following steps:
step S11: and acquiring commodity retrieval data input by the target enterprise.
In the embodiment, firstly, commodity retrieval data input by a target enterprise for commodity recommendation is acquired; wherein, the commodity retrieval data comprises, but is not limited to, commodity type, attribute, performance, commodity purchase preference and other data.
Step S12: coding the commodity retrieval data by using the trained coding model to obtain a retrieval vector; the coding model is a model obtained by training a DSSM double-tower model by utilizing historical enterprise data and commodity purchase labels; the DSSM double-tower model comprises a first tower model and a second tower model, each tower model comprises a dynamic attention mechanism layer and a plurality of fully-connected layers, and the dynamic attention mechanism layer adopts a SENet structure.
In this embodiment, after obtaining commodity retrieval data input by a target enterprise, inputting the commodity retrieval data into a trained coding model, so as to code the commodity retrieval data into a vector through the trained coding model, thereby obtaining a corresponding retrieval vector. It should be noted that, the coding model is a model obtained by training a DSSM (Deep Structured Semantic Model, deep semantic matching model) double-tower model by using historical enterprise data and commodity purchase labels; the DSSM double-tower model comprises two tower models, namely a first tower model and a second tower model, each tower model comprises a dynamic attention mechanism layer and a plurality of full-connection layers, and the dynamic attention mechanism layer adopts a SENet (squeize-and-Excitation Networks) structure.
In this embodiment, the process of creating the coding model may specifically include: collecting static data and dynamic behavior data of different enterprises to obtain historical enterprise data; constructing a feature set related to a user recommendation scene based on the historical enterprise data to form a user portrait feature library; acquiring historical commodity purchase records of different enterprises, and constructing positive and negative sample labels based on the historical commodity purchase records to obtain commodity purchase labels; inputting the commodity purchase label and the data in the user portrait feature library into a DSSM double-tower model for training for multiple times to obtain a coding model; and coding each characteristic data in the user portrait characteristic library by using the coding model to obtain a coding vector set. In this embodiment, referring to fig. 2, static data and dynamic behavior data of different enterprises are collected first to obtain historical enterprise data; the static data comprises, but is not limited to, basic information of an enterprise, industry type, enterprise scale, registered funds, participants, membership time, membership level and the like, and the dynamic behavior data comprises, but is not limited to, data such as the number of times and amount of products purchased by the enterprise, and information of curriculum looking, clicking, browsing, calling and following, consultation and communication, and functional activity related to product service of the best contact of the enterprise; then, constructing a feature set related to the user recommendation scene based on the historical enterprise data so as to form a user portrait feature library; further, historical commodity purchase records of different enterprises are obtained, positive and negative sample labels are constructed based on the historical commodity purchase records, and batch commodity purchase labels are obtained, in a specific embodiment, positive and negative sample labels can be constructed based on the historical commodity purchase records in a mode of longest public substring (Longest Common Substring), for example, as shown in fig. 3, 8 commodities are purchased by an enterprise a and an enterprise B respectively according to the commodity purchase records; the commodity 3, the commodity 4 and the commodity 5 of the upper part are intersections purchased by two enterprises together, the commodity 3 and the commodity 8 of the lower part are intersections purchased by two enterprises, the longest public substring of the upper part is 3, the longest public substring of the lower part is 2, the intersection ratio of the upper part is 3/8, and the intersection ratio of the lower part is 2/8. It should be noted that when constructing the positive and negative sample labels, the positive and negative sample labels may be constructed based on a preset business similarity rule, for example, when the business similarity rule is that the number of commodities is greater than 10 and the intersection ratio of the longest common substring is greater than 1/2, the two are judged to be similar, and the label is marked as 1, that is, the positive sample is considered; for the construction of negative samples, dissimilar enterprises can be selected as far as possible, for example, when the number of commodities is larger than 10 and the intersection ratio of the longest common substrings is smaller than 1/5, the two are judged to be dissimilar, and the two are marked as 0, namely, negative samples are considered, through the construction mode, batched positive and negative sample labels can be obtained, the formats of single samples can be 'A characteristics, B characteristics and labels 1-0', and the sample labels in the format can be used as training data sets to be input into a DSSM double-tower model for model training. Specifically, the commodity purchase label and the data in the user portrait feature library are used as training data sets to train the DSSM double-tower model for multiple times, and then a trained coding model is obtained. It should be noted that, referring to fig. 4, the DSSM twin-tower model proposed in the present application is different from the conventional DSSM twin-tower model, in which two tower models in the conventional DSSM twin-tower model are vector inputs of a user and a commodity as two tower models, and then commodity vector recall is performed, that is, features of the two tower models are different; in addition, in order to avoid model overfitting and solve the problem of limitation of manual setting of key weights, each tower model in the DSSM double-tower model provided by the application comprises a dynamic attention mechanism layer and a plurality of full-connection layers, and the dynamic attention mechanism layer adopts a SENet structure, wherein the SENet structure can learn each feature data in an input user portrait feature library to obtain sigmod probability, and the sigmod probability can be used as the weight of the dynamic attention mechanism layer in the corresponding tower model. Furthermore, the coding model obtained after training is used as a coding tool of enterprise characteristics, and all characteristic data in the input user portrait characteristic library are coded into vectors to obtain a coded vector set which is used as the input of a subsequent similarity retrieval process.
Specifically, the constructing a feature set related to the user recommendation scene based on the historical enterprise data to form a user portrait feature library may include: processing the historical enterprise data to obtain processed enterprise data, and extracting characteristics of the processed enterprise data to obtain enterprise characteristics; and constructing a feature set related to the user recommendation scene based on the enterprise features to form a user portrait feature library. Referring to fig. 2, data processing, such as data cleaning and screening, can be performed on historical enterprise data, then feature extraction is performed on the processed data by using feature engineering means such as feature extraction and feature conversion, and then feature sets related to user recommendation scenes are constructed based on the feature data and some business rule patterns defined on business, so that a database of user recommended commodities, namely a user portrait feature library, is formed.
It is noted that the model parameters of the DSSM double-tower model may be optimized during model training using a binary Cross Entropy Loss (Cross-Entropy Loss) function. Specifically, referring to fig. 4, data (including feature data corresponding to static data and dynamic behavior data) in a user portrait feature library, namely, an a feature and a B feature, are respectively input into two "tower" coding layers, and are divided into a "left tower" and a "right tower", and it is noted that, the two towers have the same structure, and the two towers sequentially input to output: the dynamic attention mechanism layer and the multi-layer full-connection layer output an A characteristic coding vector and a B characteristic coding vector after passing through the tower model, then the two coding vectors are combined by utilizing a sigmod function, so that the prediction probability is calculated, and the prediction probability output by the sigmod and the real commodity purchase label 1-0 are input into a binary cross entropy loss function to calculate the error between the prediction probability and the real label. Through the continuous iterative learning of the binary cross entropy loss function, a parameter space with the lowest loss as possible can be finally obtained, and the parameter space is the final model.
The calculation formula of the Sigmod function is as follows:
where x is the probability that the input predictive label is 1.
The calculation formula of the binary cross entropy loss function is as follows:
where o is the probability of prediction, i is the index of multi-label prediction, t is the true label, and n is the number of multi-label categories.
Specifically, each tower model in the DSSM double-tower model provided by the present application includes a dynamic attention mechanism layer and a multi-layer full-connection layer, where the dynamic attention mechanism layer adopts a SENET structure, as shown in fig. 5, the SENET structure firstly compresses and pools multiple channels of input X to obtain a vector C (i.e. Fsq), then the vector C passes through the full-connection layer of 1-2 layers (i.e. Fex), each element in the vector C is multiplied by an original channel, finally, a weight ranking after weight (i.e. Freweight) is reset is obtained, in the above pre-training manner, the importance of features can be obtained, and the importance of learned features is different according to the difference of input data of each batch, so that the feature importance can be used as dynamic weighting, and the dynamic attention mechanism layer can dynamically adjust the importance of features according to the characteristics of the input data, thereby improving the generalization capability of the DSSM double-tower model and enhancing the interpretability. The output computational expression is:
s＝σ(W[e 1 ，...，e m ]+b)；
Wherein,for the conversion of the sigmod function, e is a different feature, W is a set of weights, b is a bias, and s is an output vector.
Specifically, the calculation process of the multi-layer full connection layer in the tower model is as follows:
firstly, initializing input x by initializing weight of an input layer to obtain a vector l 1 ：
l 1 ＝W 1 x；
Then, starting from the second layer, the calculation is performed according to the following calculation formula:
l i ＝f(W i l i-1 +b i )，i＝2，....N-1；
wherein W is a weight, b is a bias;
finally, the output layer outputs y:
y＝f(W N l N-1 +b N )；
wherein, f function is an activation function than, and the expression is:
step S13: and respectively calculating the similarity of the retrieval vector and each code vector in a code vector set obtained by coding the historical enterprise data by the code model in advance, and obtaining a similarity value.
In this embodiment, after the commodity search data is encoded to obtain a search vector, further, the similarity between the search vector and each encoding vector in the encoding vector set obtained by encoding the historical enterprise data in advance using the encoding model is calculated, that is, the similarity between two encoding vectors is calculated, so as to obtain a plurality of similarity values.
In a specific embodiment, referring to FIG. 2, the computing of the search vectors and the pre-encoding of the historical enterprise data using the encoding model are performed separatelyThe similarity of each code vector in the code vector set obtained after the code to obtain a similarity value may specifically include: inputting the search vector and a coding vector set obtained by coding the historical enterprise data by the coding model in advance to Ha s h is similar to the search model, so that the similarity of the search vector and each coding vector in the coding vector set is calculated by using the cosine distance respectively, namely, the cosine distance between the two vectors is calculated, and further, a similarity value (namely, the cosine value) is obtained, and the closer the cosine value is to 1, the closer the included angle is to 0 degree, namely, the more similar the two vectors are. That is, the similarity between the search vector and each coding vector in the coding vector set is calculated by using a Hash (Hash) similarity search model, and a specific calculation formula is as follows:
wherein Q is a search vector, and D is a target vector; y is Q For the encoded search vector, y D Is the coded target vector, i.e. the coded vector in the set of coded vectors.
Step S14: and sequencing the similarity values to obtain sequenced similarity, and acquiring a preset number of similarity values from the sequenced similarity to obtain an enterprise set similar to the target enterprise.
In this embodiment, the similarity of each encoding vector in the search vector and the encoding vector set is calculated, after similarity values are obtained, all the similarity values are ranked from large to small to obtain ranked similarity, and then a preset number of similarity values are obtained from the ranked similarity, so that an enterprise set similar to the target enterprise is obtained. It can be understood that the larger the similarity value is, the higher the similarity degree is indicated, so when a preset number of similarity values are selected, the similarity values in the ordered similarity can be selected according to a front-to-back sequence, for example, the front N (topN) similarity values are selected, where N can be selected according to the actual application requirement, and then the enterprises corresponding to the topN similarity values are put into the same set, so as to obtain an enterprise set similar to the target enterprise.
Step S15: counting the purchase frequency of the commodities purchased by each enterprise in the enterprise set to obtain a frequency counting result, and sorting the frequency counting result to obtain the sorted purchase frequency.
In this embodiment, after obtaining an enterprise set similar to the target enterprise, summarizing and counting purchased commodities of each enterprise in the enterprise set, counting the frequency of purchased commodities to obtain a frequency counting result, and then sorting the frequency counting results according to the order from large to small to obtain the frequency of purchase after sorting.
Step S16: and acquiring the preset number of frequency statistics results from the ordered purchase frequency, and recommending commodities corresponding to the preset number of frequency statistics results to the target enterprise.
In this embodiment, after the frequency statistics results are ordered to obtain ordered purchase frequencies, top N (topN) frequency statistics results are obtained from the ordered purchase frequencies, and then commodities corresponding to the top N (topN) frequency statistics results are recommended to the target enterprise, so that the whole commodity recommendation process is completed.
Therefore, in the embodiment of the application, commodity retrieval data input by a target enterprise are firstly obtained, and then the trained coding model is utilized to code the commodity retrieval data to obtain a retrieval vector; the coding model is a model obtained by training a DSSM double-tower model by utilizing historical enterprise data and commodity purchase labels; the DSSM double-tower model comprises a first tower model and a second tower model, each tower model comprises a dynamic attention mechanism layer and a plurality of fully-connected layers, and the dynamic attention mechanism layer adopts a SENet structure; and then, respectively calculating the search vectors and the similarity of each code vector in a code vector set obtained by coding the historical enterprise data by utilizing the coding model in advance to obtain a similarity value, sequencing the similarity values to obtain sequenced similarity, acquiring a preset number of similarity values from the sequenced similarity to obtain an enterprise set similar to the target enterprise, finally counting the purchase frequency of commodities purchased by each enterprise in the enterprise set to obtain a frequency counting result, sequencing the frequency counting result to obtain sequenced purchase frequency, acquiring the preset number of frequency counting results from the sequenced purchase frequency, and recommending the commodities corresponding to the preset number of frequency counting results to the target enterprise. According to the embodiment of the invention, the code model created based on the DSSM double-tower model is utilized to code the commodity retrieval data input by the target enterprise to obtain the retrieval vector, and the similarity of the retrieval vector and each code vector in the code vector set is calculated, so that the enterprise set similar to the target enterprise is obtained, and the commodity to be recommended is determined according to the purchase frequency of the commodity purchased by each enterprise in the enterprise set, so that the commodity recommendation accuracy can be improved, the experience of personnel is not relied on, and the method is suitable for scenes with fewer commodity features.
The embodiment of the application discloses a specific commodity recommendation method, which is shown in fig. 6 and comprises the following steps:
step S21: and acquiring commodity retrieval data input by the target enterprise.
Step S22: coding the commodity retrieval data by using the trained coding model to obtain a retrieval vector; the coding model is a model obtained by training a DSSM double-tower model by utilizing historical enterprise data and commodity purchase labels; the DSSM double-tower model comprises a first tower model and a second tower model, each tower model comprises a dynamic attention mechanism layer and a plurality of fully-connected layers, and the dynamic attention mechanism layer adopts a SENet structure.
Step S23: and inputting the search vector and a code vector set obtained by coding the historical enterprise data by the code model in advance into a Hash similar search model to calculate the similarity of each code vector in the search vector and the code vector set by using a cosine distance respectively so as to obtain a similarity value.
Step S24: and sequencing the similarity values to obtain sequenced similarity, and acquiring a preset number of similarity values from the sequenced similarity to obtain an enterprise set similar to the target enterprise.
Step S25: counting the purchase frequency of the commodities purchased by each enterprise in the enterprise set to obtain a frequency counting result, and sorting the frequency counting result to obtain the sorted purchase frequency.
Step S26: and acquiring the preset number of frequency statistics results from the ordered purchase frequency, and respectively determining commodities corresponding to the preset number of frequency statistics results to obtain commodities to be recommended.
In this embodiment, the top N (topN) frequency statistics results may be obtained from the ordered purchase frequencies, and then the commodities corresponding to the topN frequency statistics results may be determined respectively, so as to obtain the commodity to be recommended.
Step S27: and creating a commodity recommendation list based on all the commodities to be recommended, and recommending the commodity recommendation list to the target enterprise.
In this embodiment, an ordered list of recommended products may be created based on the products to be recommended, and then the ordered list of recommended products is sent to the target enterprise, so as to complete the whole product recommendation process.
For more specific processing procedures in steps S22 to S25, reference may be made to the corresponding contents disclosed in the foregoing embodiments, and no further description is given here.
It can be seen that, in the embodiment of the present application, the DSSM double-tower model including the dynamic attention mechanism layer and the multi-layer full connection layer encodes the commodity retrieval data to obtain a retrieval vector, calculates the similarity between the retrieval vector and the encoding vector in the encoding vector set through the Hash similarity retrieval model, then selects the top N similarity values closest to the sorted similarity values, thereby obtaining an enterprise set similar to the target enterprise, counts the purchase frequency of the commodities purchased by each enterprise in the enterprise set, sorts the frequency statistics results, acquires the top N frequency statistics results from the purchase frequency after sorting, determines the commodities corresponding to the top N frequency statistics results, and finally sends the commodities to be recommended to the target enterprise in a list form. According to the embodiment of the invention, the DSSM double-tower model with the characteristics and dimensions consistent with each other and the double-tower model being the user tower is adopted, so that the generalization capability of the model can be improved, and the method is applicable to scenes with fewer commodity characteristics, thereby improving the commodity recommendation accuracy.
Correspondingly, the embodiment of the application also discloses a commodity recommending device, as shown in fig. 7, which comprises:
A first obtaining module 11, configured to obtain commodity retrieval data input by a target enterprise;
the coding module 12 is used for coding the commodity retrieval data by utilizing the trained coding model to obtain a retrieval vector; the coding model is a model obtained by training a DSSM double-tower model by utilizing historical enterprise data and commodity purchase labels; the DSSM double-tower model comprises a first tower model and a second tower model, each tower model comprises a dynamic attention mechanism layer and a plurality of fully-connected layers, and the dynamic attention mechanism layer adopts a SENet structure;
a similarity calculation module 13, configured to calculate similarities between the search vector and each of the encoded vectors in the set of encoded vectors obtained by encoding the historical enterprise data using the encoding model in advance, so as to obtain a similarity value;
a first sorting module 14, configured to sort the similarity values to obtain sorted similarity;
the second obtaining module 15 is configured to obtain a preset number of similarity values from the ordered similarities, so as to obtain an enterprise set similar to the target enterprise;
the statistics module 16 is configured to count the purchase frequency of the commodities purchased by each enterprise in the enterprise set, so as to obtain a frequency statistics result;
The second sorting module 17 is configured to sort the frequency statistics results to obtain a sorted purchase frequency;
a third obtaining module 18, configured to obtain the preset number of frequency statistics from the ordered purchase frequencies;
and the commodity recommending module 19 is used for recommending commodities corresponding to the preset number of frequency statistics to the target enterprise.
The specific workflow of each module may refer to the corresponding content disclosed in the foregoing embodiment, and will not be described herein.
In the embodiment of the application, the commodity retrieval data input by the target enterprise are acquired first, and then the trained coding model is utilized to code the commodity retrieval data to obtain a retrieval vector; the coding model is a model obtained by training a DSSM double-tower model by utilizing historical enterprise data and commodity purchase labels; the DSSM double-tower model comprises a first tower model and a second tower model, each tower model comprises a dynamic attention mechanism layer and a plurality of fully-connected layers, and the dynamic attention mechanism layer adopts a SENet structure; and then, respectively calculating the search vectors and the similarity of each code vector in a code vector set obtained by coding the historical enterprise data by utilizing the coding model in advance to obtain a similarity value, sequencing the similarity values to obtain sequenced similarity, acquiring a preset number of similarity values from the sequenced similarity to obtain an enterprise set similar to the target enterprise, finally counting the purchase frequency of commodities purchased by each enterprise in the enterprise set to obtain a frequency counting result, sequencing the frequency counting result to obtain sequenced purchase frequency, acquiring the preset number of frequency counting results from the sequenced purchase frequency, and recommending the commodities corresponding to the preset number of frequency counting results to the target enterprise. According to the embodiment of the invention, the code model created based on the DSSM double-tower model is utilized to code the commodity retrieval data input by the target enterprise to obtain the retrieval vector, and the similarity of the retrieval vector and each code vector in the code vector set is calculated, so that the enterprise set similar to the target enterprise is obtained, and the commodity to be recommended is determined according to the purchase frequency of the commodity purchased by each enterprise in the enterprise set, so that the commodity recommendation accuracy can be improved, the manual experience is not relied on, and the method is suitable for scenes with fewer commodity features.
In some specific embodiments, the similarity calculation module 13 may specifically include:
and the information input unit is used for inputting the search vector and a coded vector set obtained by coding the historical enterprise data by the coding model in advance into a Hash similar search model so as to calculate the similarity of each coded vector in the search vector and the coded vector set by using a cosine distance respectively and obtain a similarity value.
In some specific embodiments, before the first obtaining module 11, the method may further include:
the historical data acquisition unit is used for acquiring static data and dynamic behavior data of different enterprises to obtain historical enterprise data;
a first feature library construction unit, configured to construct a feature set related to a user recommendation scenario based on the historical enterprise data, so as to form a user portrait feature library;
the purchase record acquisition unit is used for acquiring historical commodity purchase records of different enterprises;
the first label construction unit is used for constructing positive and negative sample labels based on the historical commodity purchase records to obtain commodity purchase labels;
the model training unit is used for inputting the commodity purchase label and the data in the user portrait feature library into a DSSM double-tower model for training for multiple times to obtain a coding model;
And the coding unit is used for coding the characteristic data in the user portrait characteristic library by using the coding model to obtain a coding vector set.
In some specific embodiments, the first feature library construction unit may specifically include:
the data processing unit is used for processing the historical enterprise data to obtain processed enterprise data;
the feature extraction unit is used for extracting features of the processed enterprise data to obtain enterprise features;
and the second feature library construction unit is used for constructing a feature set related to the user recommendation scene based on the enterprise features so as to form a user portrait feature library.
In some specific embodiments, the first label building unit may specifically include:
and the second label construction unit is used for constructing positive and negative sample labels based on the historical commodity purchase record in a mode of the longest public substring to obtain commodity purchase labels.
In some specific embodiments, the commodity recommendation device may further include:
and the parameter optimization unit is used for optimizing the model parameters of the DSSM double-tower model by utilizing a binary cross entropy loss function.
In some embodiments, the commodity recommendation module 19 may specifically include:
The commodity determining unit is used for respectively determining commodities corresponding to the preset number of frequency statistics results to obtain commodities to be recommended;
a list creation unit for creating a commodity recommendation list based on all the commodities to be recommended;
and the list recommending unit is used for recommending the commodity recommending list to the target enterprise.
Further, the embodiment of the present application further discloses an electronic device, and fig. 8 is a block diagram of an electronic device 20 according to an exemplary embodiment, where the content of the figure is not to be considered as any limitation on the scope of use of the present application.
Fig. 8 is a schematic structural diagram of an electronic device 20 according to an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input output interface 25, and a communication bus 26. Wherein the memory 22 is configured to store a computer program that is loaded and executed by the processor 21 to implement the relevant steps in the merchandise recommendation method disclosed in any one of the foregoing embodiments. In addition, the electronic device 20 in the present embodiment may be specifically an electronic computer.
In this embodiment, the power supply 23 is configured to provide an operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and an external device, and the communication protocol to be followed is any communication protocol applicable to the technical solution of the present application, which is not specifically limited herein; the input/output interface 25 is used for acquiring external input data or outputting external output data, and the specific interface type thereof may be selected according to the specific application requirement, which is not limited herein.
The memory 22 may be a carrier for storing resources, such as a read-only memory, a random access memory, a magnetic disk, or an optical disk, and the resources stored thereon may include an operating system 221, a computer program 222, and the like, and the storage may be temporary storage or permanent storage.
The operating system 221 is used for managing and controlling various hardware devices on the electronic device 20 and computer programs 222, which may be Windows Server, netware, unix, linux, etc. The computer program 222 may further include a computer program that can be used to perform other specific tasks in addition to the computer program that can be used to perform the merchandise recommendation method performed by the electronic device 20 disclosed in any of the foregoing embodiments.
Further, the application also discloses a computer readable storage medium for storing a computer program; wherein the computer program when executed by the processor implements the previously disclosed merchandise recommendation method. For specific steps of the method, reference may be made to the corresponding contents disclosed in the foregoing embodiments, and no further description is given here.
In this specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, so that the same or similar parts between the embodiments are referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.
Those of skill would further appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative elements and steps are described above generally in terms of functionality in order to clearly illustrate the interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. The software modules may be disposed in Random Access Memory (RAM), memory, read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
Finally, it is further noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
The foregoing has described in detail a method, apparatus, device and storage medium for recommending goods, wherein specific examples are applied to illustrate the principles and embodiments of the present application, and the above examples are only used to help understand the method and core ideas of the present application; meanwhile, as those skilled in the art will have modifications in the specific embodiments and application scope in accordance with the ideas of the present application, the present description should not be construed as limiting the present application in view of the above.
Claims (10)
1. A commodity recommendation method, comprising:
acquiring commodity retrieval data input by a target enterprise;
coding the commodity retrieval data by using the trained coding model to obtain a retrieval vector; the coding model is a model obtained by training a DSSM double-tower model by utilizing historical enterprise data and commodity purchase labels; the DSSM double-tower model comprises a first tower model and a second tower model, each tower model comprises a dynamic attention mechanism layer and a plurality of fully-connected layers, and the dynamic attention mechanism layer adopts a SENet structure;
respectively calculating the similarity of the retrieval vector and each code vector in a code vector set obtained by coding the historical enterprise data by the code model in advance to obtain a similarity value;
Sorting the similarity values to obtain sorted similarity, and obtaining a preset number of similarity values from the sorted similarity to obtain an enterprise set similar to the target enterprise;
counting the purchase frequency of commodities purchased by each enterprise in the enterprise set to obtain a frequency counting result, and sorting the frequency counting result to obtain a sorted purchase frequency;
and acquiring the preset number of frequency statistics results from the ordered purchase frequency, and recommending commodities corresponding to the preset number of frequency statistics results to the target enterprise.
2. The commodity recommendation method according to claim 1, wherein the calculating of the similarity between the search vector and each of the set of encoded vectors obtained by encoding the historical enterprise data in advance using the encoding model, respectively, to obtain a similarity value, comprises:
and inputting the search vector and a code vector set obtained by coding the historical enterprise data by the code model in advance into a Hash similar search model to calculate the similarity of each code vector in the search vector and the code vector set by using a cosine distance respectively so as to obtain a similarity value.
3. The merchandise recommendation method according to claim 1, wherein before the acquiring the merchandise retrieval data inputted by the target enterprise, further comprising:
collecting static data and dynamic behavior data of different enterprises to obtain historical enterprise data;
constructing a feature set related to a user recommendation scene based on the historical enterprise data to form a user portrait feature library;
acquiring historical commodity purchase records of different enterprises, and constructing positive and negative sample labels based on the historical commodity purchase records to obtain commodity purchase labels;
inputting the commodity purchase label and the data in the user portrait feature library into a DSSM double-tower model for training for multiple times to obtain a coding model;
and coding each characteristic data in the user portrait characteristic library by using the coding model to obtain a coding vector set.
4. The merchandise recommendation method according to claim 3, wherein said constructing a feature set associated with a user recommendation scenario based on said historical enterprise data to form a user portrayal feature library comprises:
processing the historical enterprise data to obtain processed enterprise data, and extracting characteristics of the processed enterprise data to obtain enterprise characteristics;
And constructing a feature set related to the user recommendation scene based on the enterprise features to form a user portrait feature library.
5. The merchandise recommendation method according to claim 3, wherein said constructing positive and negative sample tags based on said historical merchandise purchase record to obtain merchandise purchase tags comprises:
and constructing positive and negative sample labels based on the historical commodity purchase records in a mode of the longest public substring to obtain commodity purchase labels.
6. The merchandise recommendation method according to claim 1, further comprising:
and optimizing model parameters of the DSSM double-tower model by using a binary cross entropy loss function.
7. The commodity recommendation method according to any one of claims 1 to 6, wherein recommending commodities corresponding to the preset number of frequency statistics to the target enterprise includes:
respectively determining commodities corresponding to the preset number of frequency statistics results to obtain commodities to be recommended;
and creating a commodity recommendation list based on all the commodities to be recommended, and recommending the commodity recommendation list to the target enterprise.
8. A commodity recommendation device, comprising:
The first acquisition module is used for acquiring commodity retrieval data input by a target enterprise;
the coding module is used for coding the commodity retrieval data by utilizing the trained coding model to obtain a retrieval vector; the coding model is a model obtained by training a DSSM double-tower model by utilizing historical enterprise data and commodity purchase labels; the DSSM double-tower model comprises a first tower model and a second tower model, each tower model comprises a dynamic attention mechanism layer and a plurality of fully-connected layers, and the dynamic attention mechanism layer adopts a SENet structure;
the similarity calculation module is used for calculating the similarity of each code vector in the code vector set obtained by respectively searching the vector and coding the historical enterprise data by the code model in advance to obtain a similarity value;
the first ordering module is used for ordering the similarity values to obtain ordered similarity;
the second acquisition module is used for acquiring a preset number of similarity values from the ordered similarity to obtain an enterprise set similar to the target enterprise;
the statistics module is used for counting the purchase frequency of the commodities purchased by each enterprise in the enterprise set to obtain a frequency statistics result;
The second ordering module is used for ordering the frequency statistics results to obtain ordered purchasing frequency;
the third acquisition module is used for acquiring the preset number of frequency statistics results from the ordered purchase frequencies;
and the commodity recommending module is used for recommending commodities corresponding to the preset number of frequency statistics to the target enterprise.
9. An electronic device comprising a processor and a memory; wherein the processor, when executing the computer program stored in the memory, implements the commodity recommendation method according to any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program; wherein the computer program when executed by a processor implements the merchandise recommendation method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311782104.8A CN117635238A (en) | 2023-12-22 | 2023-12-22 | Commodity recommendation method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311782104.8A CN117635238A (en) | 2023-12-22 | 2023-12-22 | Commodity recommendation method, device, equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117635238A true CN117635238A (en) | 2024-03-01 |
Family
ID=90016390
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311782104.8A Pending CN117635238A (en) | 2023-12-22 | 2023-12-22 | Commodity recommendation method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117635238A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118133046A (en) * | 2024-04-30 | 2024-06-04 | 江苏中天互联科技有限公司 | Industry data matching method and related equipment |
-
2023
- 2023-12-22 CN CN202311782104.8A patent/CN117635238A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118133046A (en) * | 2024-04-30 | 2024-06-04 | 江苏中天互联科技有限公司 | Industry data matching method and related equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2021203819A1 (en) | Content recommendation method and apparatus, electronic device, and storage medium | |
| CN111797321B (en) | Personalized knowledge recommendation method and system for different scenes | |
| CN115878904B (en) | Intellectual property personalized recommendation method, system and medium based on deep learning | |
| CN113033090B (en) | Push model training method, data push device and storage medium | |
| CN112989761B (en) | Text classification method and device | |
| CN106610970A (en) | Collaborative filtering-based content recommendation system and method | |
| CN117635238A (en) | Commodity recommendation method, device, equipment and storage medium | |
| CN112085525A (en) | User network purchasing behavior prediction research method based on hybrid model | |
| CN114358657B (en) | Post recommendation method and device based on model fusion | |
| CN114511058B (en) | Load element construction method and device for electric power user portrait | |
| CN116308685B (en) | Product recommendation method and system based on aspect emotion prediction and collaborative filtering | |
| CN115062732A (en) | Resource sharing cooperation recommendation method and system based on big data user tag information | |
| CN114861050A (en) | Feature fusion recommendation method and system based on neural network | |
| CN117807232A (en) | Commodity classification method, commodity classification model construction method and device | |
| CN115600017A (en) | Feature coding model training method and device and media object recommendation method and device | |
| CN114637920A (en) | Object recommendation method and device | |
| CN117194742A (en) | Industrial software component recommendation method and system | |
| CN116881409A (en) | Commodity information automatic question-answering method based on e-commerce knowledge graph | |
| CN116957128A (en) | Service index prediction method, device, equipment and storage medium | |
| CN117194966A (en) | Training method and related device for object classification model | |
| CN114022233A (en) | Novel commodity recommendation method | |
| CN114154564A (en) | Method and device for determining relevance based on heterogeneous graph, electronic equipment and storage medium | |
| Ji et al. | Tri-objective optimization-based cascade ensemble pruning for deep forest | |
| CN112200255A (en) | Information redundancy removing method for sample set | |
| CN118170993B (en) | Educational resource recommendation method based on contrast learning and field factor decomposition |
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 |