CN116993374A - Model optimization method, device, equipment and medium based on deep neural network - Google Patents
Model optimization method, device, equipment and medium based on deep neural network Download PDFInfo
- Publication number
- CN116993374A CN116993374A CN202211166655.7A CN202211166655A CN116993374A CN 116993374 A CN116993374 A CN 116993374A CN 202211166655 A CN202211166655 A CN 202211166655A CN 116993374 A CN116993374 A CN 116993374A
- Authority
- CN
- China
- Prior art keywords
- channel
- user
- product
- optimized
- vector
- 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
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 86
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000005457 optimization Methods 0.000 title claims abstract description 34
- 239000013598 vector Substances 0.000 claims abstract description 185
- 238000012549 training Methods 0.000 claims abstract description 47
- 230000008569 process Effects 0.000 claims description 15
- 238000004590 computer program Methods 0.000 claims description 12
- 238000010845 search algorithm Methods 0.000 claims description 12
- 238000003860 storage Methods 0.000 claims description 10
- 108091006146 Channels Proteins 0.000 description 314
- 101100481876 Danio rerio pbk gene Proteins 0.000 description 10
- 101100481878 Mus musculus Pbk gene Proteins 0.000 description 10
- 238000005266 casting Methods 0.000 description 8
- 238000004891 communication Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 230000006399 behavior Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000026676 system process Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000003542 behavioural effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001960 triggered effect Effects 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Finance (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- General Business, Economics & Management (AREA)
- Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Marketing (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the field of computers, and provides a model optimization method, device, equipment and medium based on a deep neural network, wherein the method comprises the following steps: determining a first user channel vector based on the user dense features, the user sparse features and the channel features, and training a deep neural network of the user channel based on the first user channel vector to obtain an optimized user channel marketing model; determining a first product channel vector based on the product density feature, the product sparse feature and the channel feature, and training a deep neural network of the product channel based on the first product channel vector to obtain an optimized product channel marketing model; and obtaining an optimized marketing model based on the optimized user channel marketing model and the optimized product channel marketing model. The model optimization method based on the deep neural network provided by the embodiment of the invention can accurately recommend the optimized combination among users, products and channels through the optimized marketing model.
Description
Technical Field
The present invention relates to the field of computers, and in particular, to a method, an apparatus, a device, and a medium for model optimization based on a deep neural network.
Background
In the current telecom operator industry, the main modeling method is to respectively establish a model through the combination of user characteristics and product characteristics and the combination of user characteristics and channel characteristics, and aggregate the model results as recommendation results of the combination of users, products and channels. However, in the modeling described above, only the user features and the channel features are combined, and no product features and channel features are combined.
Disclosure of Invention
The invention provides a model optimization method, device, equipment and medium based on a deep neural network, which aim to accurately recommend the optimal combination among users, products and channels.
In a first aspect, the present invention provides a model optimization method based on a deep neural network, including:
determining a first user channel vector based on the user dense features, the user sparse features and the channel features, and training a deep neural network of the user channel based on the first user channel vector to obtain an optimized user channel marketing model;
Determining a first product channel vector based on the product density feature, the product sparse feature and the channel feature, and training a deep neural network of a product channel based on the first product channel vector to obtain an optimized product channel marketing model;
and obtaining an optimized marketing model based on the optimized user channel marketing model and the optimized product channel marketing model.
In one embodiment, after obtaining the optimized marketing model based on the optimized user channel marketing model and the optimized product channel marketing model, the method further includes:
and storing all second user channel vectors obtained in the deep neural network training process of the user channel and all second product channel vectors obtained in the deep neural network training process of the product channel into a memory database.
The method further comprises the following steps of:
determining a marketing recommendation scene, wherein the marketing recommendation scene comprises a product-based user recommendation scene and a product-based recommendation user scene;
If the marketing recommendation scene is based on a product recommendation user scene, matching a target recommendation user based on the optimized product channel marketing model in combination with first input information;
and if the marketing recommendation scene is based on the user recommendation product scene, matching a target recommendation product based on the optimized user channel marketing model and combining with second input information.
The matching of the target recommended user based on the optimized product channel marketing model combined with the first input information comprises the following steps:
calculating a target product channel vector based on the optimized product channel marketing model and the input channel ID and product characteristics;
calculating first vector similarity between the channel vector of the target product and all second user channel vectors in the memory database, and searching out a vector with highest first similarity in the first vector similarity through an approximate nearest neighbor search algorithm;
and matching the user corresponding to the vector with the highest first similarity as a target recommended user.
The matching of the target recommended product based on the optimized user channel marketing model and the second input information comprises the following steps:
calculating a target user channel vector based on the optimized user channel marketing model and the input channel ID and user characteristics;
Calculating second vector similarity between the target user channel vector and all second product channel vectors in the memory database, and searching out a vector with highest second similarity in the second vector similarity through an approximate nearest neighbor search algorithm;
and matching the product corresponding to the vector with the highest second similarity as a target recommended product.
The determining a first user channel vector based on the user dense features, the user sparse features, and the channel features comprises:
normalizing the user dense features to obtain normalized user dense features, and performing single-heat encoding on the normalized user dense features to generate a first user channel vector to be processed;
performing single-heat coding on the user sparse features and the channel features to generate a second user channel vector to be processed;
and connecting the first to-be-processed user channel vector with the second to-be-processed user channel vector to obtain the first user channel vector.
In a second aspect, the present invention provides a model optimization apparatus based on a deep neural network, including:
the first training module is used for determining a first user channel vector based on the user dense features, the user sparse features and the channel features, training a deep neural network of the user channel based on the first user channel vector, and obtaining an optimized user channel marketing model;
The second training module is used for determining a first product channel vector based on the product density characteristic, the product sparse characteristic and the channel characteristic, training a deep neural network of the product channel based on the first product channel vector, and obtaining an optimized product channel marketing model;
and the determining module is used for obtaining the optimized marketing model based on the optimized user channel marketing model and the optimized product channel marketing model.
In one embodiment, the model optimization device based on the deep neural network further comprises a matching module, wherein the matching module is used for:
determining a marketing recommendation scene, wherein the marketing recommendation scene comprises a product-based user recommendation scene and a product-based recommendation user scene;
if the marketing recommendation scene is based on a product recommendation user scene, matching a target recommendation user based on the optimized product channel marketing model in combination with first input information;
and if the marketing recommendation scene is based on the user recommendation product scene, matching a target recommendation product based on the optimized user channel marketing model and combining with second input information.
In a third aspect, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the deep neural network-based model optimization method of the first aspect when executing the program.
In a fourth aspect, the present invention also provides a non-transitory computer readable storage medium comprising a computer program which, when executed by the processor, implements the deep neural network based model optimization method of the first aspect.
In a fifth aspect, the present invention also provides a computer program product comprising a computer program which, when executed by the processor, implements the deep neural network based model optimization method of the first aspect.
According to the model optimization method, device, equipment and medium based on the deep neural network, the first user channel vector is determined based on the user dense features, the user sparse features and the channel features, and the deep neural network of the user channel is trained based on the first user channel vector, so that an optimized user channel marketing model is obtained; determining a first product channel vector based on the product density feature, the product sparse feature and the channel feature, and training a deep neural network of the product channel based on the first product channel vector to obtain an optimized product channel marketing model; and obtaining an optimized marketing model based on the optimized user channel marketing model and the optimized product channel marketing model.
In the model optimization process based on the deep neural network, an optimized user channel marketing model is obtained through user feature and channel feature training, and an optimized channel marketing model is obtained through product feature and channel feature training, so that not only channel features and user features but also channel features and product features are combined in the optimized marketing model, three elements of users, products and channels are integrated into the model uniformly, complex user, product and channel interaction features can be captured based on nonlinear transformation, and the optimized combination among users, products and channels can be accurately recommended through the optimized marketing model.
Drawings
In order to more clearly illustrate the technical solutions of the present invention, the following description will be given with a brief introduction to the drawings used in the embodiments or the description of the prior art, it being obvious that the drawings in the following description are some embodiments of the present invention, and that other drawings can be obtained from these drawings without the inventive effort of a person skilled in the art.
FIG. 1 is a flow chart of a model optimization method based on a deep neural network provided by the invention;
FIG. 2 is a schematic structural diagram of a model optimizing device based on a deep neural network;
fig. 3 is a schematic structural diagram of an electronic device provided by the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Further, the method, the device, the equipment and the medium for optimizing the model based on the deep neural network are described with reference to fig. 1 to 3. FIG. 1 is a flow chart of a model optimization method based on a deep neural network provided by the invention; FIG. 2 is a schematic structural diagram of a model optimizing device based on a deep neural network; fig. 3 is a schematic structural diagram of an electronic device provided by the present invention.
The embodiments of the present invention provide embodiments of a deep neural network based model optimization method, it should be noted that although a logic sequence is shown in the flowchart, the steps shown or described may be accomplished in a different order than that shown or described herein under certain data.
The embodiment of the invention takes the electronic equipment as an execution main body for example, and takes the marketing system as one of the expression forms of the electronic equipment, and the embodiment of the invention is not limited.
Referring to fig. 1, fig. 1 is a schematic flow chart of a model optimization method based on a deep neural network provided by the invention. The model optimization method based on the deep neural network provided by the embodiment of the invention comprises the following steps:
step 101, determining a first user channel vector based on a user dense feature, a user sparse feature and a channel feature, and training a deep neural network of a user channel based on the first user channel vector to obtain an optimized user channel marketing model;
step 102, determining a first product channel vector based on the product density feature, the product sparse feature and the channel feature, and training a deep neural network of a product channel based on the first product channel vector to obtain an optimized product channel marketing model;
and step 103, obtaining an optimized marketing model based on the optimized user channel marketing model and the optimized product channel marketing model.
In order to improve the quality of model training, the data preparation in the early stage is needed, and the specific process is as follows:
Obtaining original data: user attribute data, user behavior data, contact feedback data, product attribute data, and channel data are obtained from the system as raw data inputs. The original data is subjected to certain cleaning and pretreatment for further modeling;
constructing a sample: depending on the business objectives, care should be taken to avoid sample selection bias issues (Sample Selection Bias) when constructing how to divide positive and negative samples. For example, if the business objective is to promote the order rate, the positive sample can be set as the product which is contacted and processed by the user, and the negative sample can be set as the negative sample randomly selected in the batch with a certain proportion of the product which is contacted and not processed by the user to avoid sample selection deviation;
defining sample characteristics: each sample should include user characteristics, product characteristics, channel characteristics, which may include user ID, user base attributes, user historical subscription services, consumption value, behavioral preferences, dynamic status, etc.; product characteristics may include product ID, product base attributes, ordered conditions, etc.; the channel characteristics include a channel ID.
It should be further noted that the present invention is an improvement based on the original double-tower model, and incorporates channel characteristics as a third dimension into the model. The two tower models in the double tower model are respectively composed of user characteristics, channel characteristics and product characteristics, and the channel characteristics are used as common input design of the two tower models, so that the cross characteristics of the channel, the user and the product can be captured, and the generalization of the model is improved.
Features can be classified into dense features (dense features) and sparse features (sparse features). Dense features such as traffic usage, number of calls, ARPU, etc. numeric features. Sparse features such as ID, region, gender, subscribed services, etc.
Further, the marketing system processes the user dense features, the user sparse features and the channel features to obtain a first user channel marketing vector. Further, the marketing system trains a Deep neural network (Deep-Learning Neural Network, DNN) of the user channel through the first user channel marketing vector to obtain an optimized user channel marketing model, wherein the Deep neural network DNN of the user channel consists of a fully connected layer with 256 input dimensions, a fully connected layer with 128 input dimensions and a fully connected layer with 32 output dimensions. The optimized user channel marketing model may also be an optimized user channel tower model.
Further, the marketing system processes the product density features, the product sparsity features and the channel features to obtain a first product channel marketing vector. Further, the marketing system trains the deep neural network of the product channel through the first product channel marketing vector to obtain an optimized product channel marketing model, wherein the deep neural network of the product channel consists of a fully connected layer with an input dimension of 128, a fully connected layer with an input dimension of 64 and a fully connected layer with an output dimension of 32. The optimized product channel marketing model may also become an optimized product channel power model.
Further, the marketing system determines the optimized user channel marketing model and the optimized user channel marketing model as marketing models for scene matching, namely, the optimized user channel power model and the optimized product channel power model as marketing double-tower power models for scene matching.
It should be further noted that, the optimized user channel marketing model is mainly used for extracting the user channel marketing vector, the optimized product channel marketing model is mainly used for extracting the product channel marketing vector, and the dimensions of the extracted results of the optimized user channel marketing model and the optimized product channel marketing model are the same, that is, the network layer number and dimensions inside the optimized user channel marketing model and the optimized product channel marketing model can be different, but the output dimensions must be the same.
Therefore, the optimized user channel marketing model and the optimized product channel marketing model can be understood as feature extractors, and the selection of the feature extractors can be customized, such as structures of a multi-layer perceptron (Multilayer Perceptron), a Convolutional Neural Network (CNN), a long-short-term memory model (LSTM), a transformer and the like. The invention uses a multi-layer perceptron with 3 hidden layers, wherein the hidden layers are all fully connected layers (fully connected layers), one perceptron has a plurality of inputs and an output, the inputs and the output are linear relations obtained by model training, and the output is input to the next layer through a nonlinear activation function.
According to the model optimization method based on the deep neural network, a first user channel vector is determined based on the user dense features, the user sparse features and the channel features, and the deep neural network of the user channel is trained based on the first user channel vector to obtain an optimized user channel marketing model; determining a first product channel vector based on the product density feature, the product sparse feature and the channel feature, and training a deep neural network of the product channel based on the first product channel vector to obtain an optimized product channel marketing model; and obtaining an optimized marketing model based on the optimized user channel marketing model and the optimized product channel marketing model.
In the model optimization process based on the deep neural network, an optimized user channel marketing model is obtained through user feature and channel feature training, and an optimized channel marketing model is obtained through product feature and channel feature training, so that not only channel features and user features but also channel features and product features are combined in the optimized marketing model, three elements of users, products and channels are integrated into the model uniformly, complex user, product and channel interaction features can be captured based on nonlinear transformation, and the optimized combination among users, products and channels can be accurately recommended through the optimized marketing model.
Further, based on the user dense features, the user sparse features, and the channel features described in step 101, a specific analysis for determining the first user channel vector is as follows:
normalizing the user dense features to obtain normalized user dense features, and performing single-heat encoding on the normalized user dense features to generate a first user channel vector to be processed;
performing single-heat coding on the user sparse features and the channel features to generate a second user channel vector to be processed;
and connecting the first to-be-processed user channel vector with the second to-be-processed user channel vector to obtain the first user channel vector.
Specifically, for user features: the marketing system firstly carries out normalization processing on the user dense features (user dense feature) to obtain normalized user dense features, wherein the formula of normalization processing is z= (X-mu)/sigma, wherein X is a sample value, mu is a sample mean value, and sigma is a sample standard deviation. Further, the marketing system carries out one-hot one-time encoding on the normalized user dense features to generate a first waiting user channel marketing vector.
Further, the marketing system carries out one-hot single-heat coding on the sparse feature (user sparse feature) and the channel feature (channel feature) of the user to generate a second waiting user channel mapping vector, and the second waiting user channel mapping vector is obtained after one-hot single-heat coding, wherein the marked value is 1, and the other values are 0.
Further, the marketing system connects the first waiting user channel sounding vector and the second waiting user channel sounding vector through a connection to obtain a first user channel sounding vector, so that the first user channel sounding vector is expressed as v user+channel ＝concat(v user dense feature ,v user spars efeature ,v channel feature )。
Further, for product characteristics: the marketing system normalizes the product density features (item dense feature) to obtain normalized product density features. Further, the marketing system carries out one-hot one-time encoding on the normalized product density characteristics to generate a first product channel waiting vector.
Further, the marketing system performs one-hot independent encoding on the product sparse feature (item sparse feature) and the channel feature (channel feature) to generate a second product channel marketing vector to be processed.
Further, the marketing system connects the first product channel enabling vector to be processed and the second product channel enabling vector to be processed through a concatate to obtain the first product channel enabling vector, and therefore the first product channel enabling vector is expressed as v item+channel ＝concat(v item dense feature ,v item sparse feature ,v channel feature )。
According to the embodiment of the invention, the channel characteristics are taken as the third dimension to be incorporated into the model to form the user characteristics, the channel characteristics, the product characteristics and the channel characteristics, so that the cross characteristics of the channel, the user and the product are captured, and the generalization of the trained model is improved.
Further, after the optimized marketing model is obtained based on the optimized user channel marketing model and the optimized product channel marketing model in step 103, the vector in the training process needs to be stored, and the specific analysis is as follows:
and storing all second user channel vectors obtained in the deep neural network training process of the user channel and all second product channel vectors obtained in the deep neural network training process of the product channel into a memory database.
Specifically, during the training of the deep neural network DNN of the user channel by the first user channel sounding vector, a second user channel sounding vector (user+ channel embedding) of 32 dimensions is continuously output. Meanwhile, in the training process of the deep neural network of the product channel through the first product channel sounding vector, a second product channel vector (item+ channel embedding) with 32 dimensions is also continuously output.
Therefore, the marketing system needs to continuously store all second user channel enabling vectors output in the process of the deep neural network DNN training of the user channels and all second product channel vectors output in the process of the deep neural network training of the product channels into an in-memory database (Key-Value database) for real-time online matching.
The embodiment of the invention realizes real-time online matching through the memory database.
Further, after the optimized marketing model is obtained based on the optimized user channel marketing model and the optimized product channel marketing model in step 103, on-line matching can be performed in real time after the optimized marketing model is needed to be matched, and the specific analysis is as follows:
determining a marketing recommendation scene, wherein the marketing recommendation scene comprises a product-based user recommendation scene and a product-based recommendation user scene;
if the marketing recommendation scene is based on a product recommendation user scene, matching a target recommendation user based on the optimized product channel marketing model in combination with first input information;
and if the marketing recommendation scene is based on the user recommendation product scene, matching a target recommendation product based on the optimized user channel marketing model and combining with second input information.
Specifically, when products match or users match, a marketing recommendation scene needs to be determined, wherein the marketing recommendation scene comprises a product-based recommendation scene and a product-based recommendation user scene. Based on the product recommendation user scenario, i.e. the scenario for "product find customer base" (the product to be marketed has been determined, the best matching customer base needs to be found). The product scenario is recommended based on the user, i.e. for a "user personalized recommendation" scenario (for each user, find the best matching product).
Further, if the marketing recommendation scene is determined to be based on the product recommendation user scene, the marketing system determines the input first input information and matches the target recommendation user to be recommended by combining the optimized product channel marketing model with the first input information.
Further, if the marketing recommendation scene is determined to be based on the user recommendation product scene, the marketing system determines the second input information, and matches the target recommendation product to be recommended by combining the optimized user channel marketing model with the second input information.
According to the embodiment of the invention, the optimized product channel marketing model and the optimized user channel marketing model have no mutual dependency relationship, and the optimized product channel marketing model and the optimized user channel marketing model can be independently matched with target recommended users or target recommended products, so that real-time rapid matching marketing plan is realized.
Further, based on the optimized product channel marketing model combined with the first input information, the specific analysis of the matching target recommended user is as follows:
calculating a target product channel vector based on the optimized product channel marketing model and the input channel ID and product characteristics;
Calculating first vector similarity between the channel vector of the target product and all second user channel vectors in the memory database, and searching out a vector with highest first similarity in the first vector similarity through an approximate nearest neighbor search algorithm;
and matching the user corresponding to the vector with the highest first similarity as a target recommended user.
Specifically, a user scenario is recommended based on a product: namely, the scenes of the product finding group are used for determining reachable marketing channels, and the channel ID and product characteristics are used as input information for each channel and are input into the optimized product channel marketing model. And calculating a target product channel marketing vector of channel ID+product characteristics through the optimized product channel marketing model.
Further, the marketing system calculates the first vector cosine similarity between the target product channel casting vector and all the second user channel casting vectors in the memory database in a cosine similarity (i.e. L2 regularization of the inner product) mode.
The essence of cosine similarity is that a search graph is constructed by using Euclidean distance, so that the transmissibility of distance comparison is maintained, the consistency of results of offline training and online searching of a model is ensured, and the model effect can be improved.
Further, the marketing system searches the topK casting vector with the highest first similarity in the first vector cosine similarity through an approximate nearest neighbor search algorithm, wherein the topK casting vector with the highest first similarity is not only one vector, but also a plurality of vectors. Further, the marketing system matches the user corresponding to the topK sounding vector with the highest similarity as the target recommended user, namely, matches the user group corresponding to the topK sounding vector with the highest similarity as the target recommended user.
It should be noted that, the HNSW (Hierarchical Navigable Small World graphs) search algorithm in the approximate nearest neighbor search algorithm (Approximate Nearest Neighbor, ANN) in the present invention can realize rapid matching of mass data. HNSW is a graph-based stored data structure, and the general idea is as follows: in layer=0 layers, all points in the connected graph are included. As the number of layers increases, the number of points per layer gradually decreases and follows the law of exponential decay. The maximum number of layers of the graph nodes is determined by a random exponential probability decay function. The node exists in all layers down from the highest layer where the node is located. When the HNSW is queried, the search starts from the highest layer.
According to the embodiment of the invention, the topK marketing vector with the highest similarity can be quickly searched by combining the optimized product channel marketing model with the approximate nearest neighbor searching algorithm, so that a target recommended user can be quickly matched in a large scale.
Further, based on the optimized user channel marketing model and the second input information, the specific analysis of the matched target recommended product is as follows:
calculating a target user channel vector based on the optimized user channel marketing model and the input channel ID and user characteristics;
calculating second vector similarity between the target user channel vector and all second product channel vectors in the memory database, and searching out a vector with highest second similarity in the second vector similarity through an approximate nearest neighbor search algorithm;
and matching the product corresponding to the vector with the highest second similarity as a target recommended product.
Specifically, based on the user recommended product scenario: the user personalized recommendation scene is used for determining reachable marketing channels, and the channel ID and the user characteristics are used as input information for each channel and are input into the optimized user channel marketing model. And calculating a target user channel marketing vector of channel ID+user characteristics through the optimized user channel marketing model. It should be noted that in this scenario, the user features may include user real-time behavior data such as browsing, subscription, push feedback, and position change, so that when the recommendation is triggered, the best matching product may be calculated in real time according to the current state of the user for recommendation.
Further, the marketing system calculates second vector cosine similarity between the target user channel casting vector and all second product channel casting vectors in the memory database in a cosine similarity mode.
Further, the marketing system searches the topK casting vector with the highest second similarity in the second vector cosine similarity through an approximate nearest neighbor search algorithm, wherein the topK casting vector with the highest second similarity is not only one vector, but also a plurality of vectors. Further, the marketing system matches the product corresponding to the topK sounding vector with the highest second similarity as the target recommended product, namely, matches the product group corresponding to the topK sounding vector with the highest second similarity as the target recommended product.
According to the embodiment of the invention, the topK marketing vector with the highest similarity can be quickly searched by combining the optimized user channel marketing model with the approximate nearest neighbor searching algorithm, so that a target recommended product can be quickly matched in a large scale.
Further, the model optimization device based on the deep neural network provided by the invention and the model optimization method based on the deep neural network provided by the invention are correspondingly referred to each other.
Fig. 2 is a schematic structural diagram of a model optimizing device based on a deep neural network according to the present invention, where the model optimizing device based on the deep neural network includes:
the first training module 201 is configured to determine a first user channel vector based on the user dense feature, the user sparse feature and the channel feature, and train the deep neural network of the user channel based on the first user channel vector to obtain an optimized user channel marketing model;
the second training module 202 is configured to determine a first product channel vector based on the product dense feature, the product sparse feature and the channel feature, and train the deep neural network of the product channel based on the first product channel vector to obtain an optimized product channel marketing model;
and the determining module 203 is configured to obtain an optimized marketing model based on the optimized user channel marketing model and the optimized product channel marketing model.
Further, the first training module 201 is further configured to:
normalizing the user dense features to obtain normalized user dense features, and performing single-heat encoding on the normalized user dense features to generate a first user channel vector to be processed;
Performing single-heat coding on the user sparse features and the channel features to generate a second user channel vector to be processed;
and connecting the first to-be-processed user channel vector with the second to-be-processed user channel vector to obtain the first user channel vector.
Further, the model optimizing device based on the deep neural network further comprises a storage module, wherein the storage module is used for:
and storing all second user channel vectors obtained in the deep neural network training process of the user channel and all second product channel vectors obtained in the deep neural network training process of the product channel into a memory database.
Further, the model optimizing device based on the deep neural network further comprises a matching module, wherein the matching module is used for:
determining a marketing recommendation scene, wherein the marketing recommendation scene comprises a product-based user recommendation scene and a product-based recommendation user scene;
if the marketing recommendation scene is based on a product recommendation user scene, matching a target recommendation user based on the optimized product channel marketing model in combination with first input information;
and if the marketing recommendation scene is based on the user recommendation product scene, matching a target recommendation product based on the optimized user channel marketing model and combining with second input information.
Further, the matching module is further configured to:
calculating a target product channel vector based on the optimized product channel marketing model and the input channel ID and product characteristics;
calculating first vector similarity between the channel vector of the target product and all second user channel vectors in the memory database, and searching out a vector with highest first similarity in the first vector similarity through an approximate nearest neighbor search algorithm;
and matching the user corresponding to the vector with the highest first similarity as a target recommended user.
Further, the matching module is further configured to:
calculating a target user channel vector based on the optimized user channel marketing model and the input channel ID and user characteristics;
calculating second vector similarity between the target user channel vector and all second product channel vectors in the memory database, and searching out a vector with highest second similarity in the second vector similarity through an approximate nearest neighbor search algorithm;
and matching the product corresponding to the vector with the highest second similarity as a target recommended product.
The specific embodiment of the model optimizing device based on the deep neural network provided by the invention is basically the same as each embodiment of the model optimizing method based on the deep neural network, and is not repeated herein.
Fig. 3 illustrates a physical schematic diagram of an electronic device, as shown in fig. 3, the electronic device may include: processor 310, communication interface (Communications Interface) 320, memory 330 and communication bus 340, wherein processor 310, communication interface 320, memory 330 accomplish communication with each other through communication bus 340. The processor 310 may invoke logic instructions in the memory 330 to perform a deep neural network based model optimization method comprising:
determining a first user channel vector based on the user dense features, the user sparse features and the channel features, and training a deep neural network of the user channel based on the first user channel vector to obtain an optimized user channel marketing model;
determining a first product channel vector based on the product density feature, the product sparse feature and the channel feature, and training a deep neural network of a product channel based on the first product channel vector to obtain an optimized product channel marketing model;
and obtaining an optimized marketing model based on the optimized user channel marketing model and the optimized product channel marketing model.
Further, the logic instructions in the memory 330 described above may be implemented in the form of software functional units and may be stored in a computer-readable storage medium when sold or used as a stand-alone product. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
In another aspect, the present invention also provides a computer program product comprising a computer program stored on a non-transitory computer readable storage medium, the computer program comprising program instructions which, when executed by a computer, enable the computer to perform a deep neural network based model optimization method provided by the above methods, the method comprising:
Determining a first user channel vector based on the user dense features, the user sparse features and the channel features, and training a deep neural network of the user channel based on the first user channel vector to obtain an optimized user channel marketing model;
determining a first product channel vector based on the product density feature, the product sparse feature and the channel feature, and training a deep neural network of a product channel based on the first product channel vector to obtain an optimized product channel marketing model;
and obtaining an optimized marketing model based on the optimized user channel marketing model and the optimized product channel marketing model.
In yet another aspect, the present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, is implemented to perform the above provided deep neural network based model optimization method, the method comprising:
determining a first user channel vector based on the user dense features, the user sparse features and the channel features, and training a deep neural network of the user channel based on the first user channel vector to obtain an optimized user channel marketing model;
Determining a first product channel vector based on the product density feature, the product sparse feature and the channel feature, and training a deep neural network of a product channel based on the first product channel vector to obtain an optimized product channel marketing model;
and obtaining an optimized marketing model based on the optimized user channel marketing model and the optimized product channel marketing model.
The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on this understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective embodiments or some parts of the embodiments.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (10)
1. A model optimization method based on a deep neural network, comprising the steps of:
determining a first user channel vector based on the user dense features, the user sparse features and the channel features, and training a deep neural network of the user channel based on the first user channel vector to obtain an optimized user channel marketing model;
determining a first product channel vector based on the product density feature, the product sparse feature and the channel feature, and training a deep neural network of a product channel based on the first product channel vector to obtain an optimized product channel marketing model;
and obtaining an optimized marketing model based on the optimized user channel marketing model and the optimized product channel marketing model.
2. The deep neural network-based model optimization method of claim 1, wherein the obtaining the optimized marketing model based on the optimized user channel marketing model and the optimized product channel marketing model further comprises:
and storing all second user channel vectors obtained in the deep neural network training process of the user channel and all second product channel vectors obtained in the deep neural network training process of the product channel into a memory database.
3. The deep neural network-based model optimization method of claim 2, wherein the obtaining the optimized marketing model based on the optimized user channel marketing model and the optimized product channel marketing model further comprises:
determining a marketing recommendation scene, wherein the marketing recommendation scene comprises a product-based user recommendation scene and a product-based recommendation user scene;
if the marketing recommendation scene is based on a product recommendation user scene, matching a target recommendation user based on the optimized product channel marketing model in combination with first input information;
And if the marketing recommendation scene is based on the user recommendation product scene, matching a target recommendation product based on the optimized user channel marketing model and combining with second input information.
4. The model optimizing method based on the deep neural network according to claim 3, wherein the matching the target recommended user based on the optimized product channel marketing model in combination with the first input information comprises:
calculating a target product channel vector based on the optimized product channel marketing model and the input channel ID and product characteristics;
calculating first vector similarity between the channel vector of the target product and all second user channel vectors in the memory database, and searching out a vector with highest first similarity in the first vector similarity through an approximate nearest neighbor search algorithm;
and matching the user corresponding to the vector with the highest first similarity as a target recommended user.
5. The model optimizing method based on the deep neural network according to claim 3, wherein the matching the target recommended product based on the optimized user channel marketing model in combination with the second input information comprises:
Calculating a target user channel vector based on the optimized user channel marketing model and the input channel ID and user characteristics;
calculating second vector similarity between the target user channel vector and all second product channel vectors in the memory database, and searching out a vector with highest second similarity in the second vector similarity through an approximate nearest neighbor search algorithm;
and matching the product corresponding to the vector with the highest second similarity as a target recommended product.
6. The depth neural network based model optimization method of any one of claims 1-5, wherein the determining a first user channel vector based on the user dense feature, the user sparse feature, and the channel feature comprises:
normalizing the user dense features to obtain normalized user dense features, and performing single-heat encoding on the normalized user dense features to generate a first user channel vector to be processed;
performing single-heat coding on the user sparse features and the channel features to generate a second user channel vector to be processed;
and connecting the first to-be-processed user channel vector with the second to-be-processed user channel vector to obtain the first user channel vector.
7. A deep neural network-based model optimization device, comprising:
the first training module is used for determining a first user channel vector based on the user dense features, the user sparse features and the channel features, training a deep neural network of the user channel based on the first user channel vector, and obtaining an optimized user channel marketing model;
the second training module is used for determining a first product channel vector based on the product density characteristic, the product sparse characteristic and the channel characteristic, training a deep neural network of the product channel based on the first product channel vector, and obtaining an optimized product channel marketing model;
and the determining module is used for obtaining the optimized marketing model based on the optimized user channel marketing model and the optimized product channel marketing model.
8. The deep neural network-based model optimization device of claim 7, further comprising a matching module for:
determining a marketing recommendation scene, wherein the marketing recommendation scene comprises a product-based user recommendation scene and a product-based recommendation user scene;
If the marketing recommendation scene is based on a product recommendation user scene, matching a target recommendation user based on the optimized product channel marketing model in combination with first input information;
and if the marketing recommendation scene is based on the user recommendation product scene, matching a target recommendation product based on the optimized user channel marketing model and combining with second input information.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the deep neural network based model optimization method of any one of claims 1 to 6 when executing the computer program.
10. A non-transitory computer readable storage medium comprising a computer program, characterized in that the computer program when executed by a processor implements the deep neural network based model optimization method of any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211166655.7A CN116993374A (en) | 2022-09-23 | 2022-09-23 | Model optimization method, device, equipment and medium based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211166655.7A CN116993374A (en) | 2022-09-23 | 2022-09-23 | Model optimization method, device, equipment and medium based on deep neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116993374A true CN116993374A (en) | 2023-11-03 |
Family
ID=88523816
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211166655.7A Pending CN116993374A (en) | 2022-09-23 | 2022-09-23 | Model optimization method, device, equipment and medium based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116993374A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117407727A (en) * | 2023-11-28 | 2024-01-16 | 星环信息科技(上海)股份有限公司 | Vector similarity determining method and vector searching method |
-
2022
- 2022-09-23 CN CN202211166655.7A patent/CN116993374A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117407727A (en) * | 2023-11-28 | 2024-01-16 | 星环信息科技(上海)股份有限公司 | Vector similarity determining method and vector searching method |
| CN117407727B (en) * | 2023-11-28 | 2024-05-14 | 星环信息科技(上海)股份有限公司 | Vector similarity determining method and vector searching method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116261731A (en) | Relation learning method and system based on multi-hop attention-seeking neural network | |
| CN112131890A (en) | Method, device and equipment for constructing intelligent recognition model of conversation intention | |
| WO2022252455A1 (en) | Methods and systems for training graph neural network using supervised contrastive learning | |
| CN113268667B (en) | Chinese comment emotion guidance-based sequence recommendation method and system | |
| Ouhbi et al. | Deep learning based recommender systems | |
| CN110737730A (en) | Unsupervised learning-based user classification method, unsupervised learning-based user classification device, unsupervised learning-based user classification equipment and storage medium | |
| CN110299194B (en) | Similar case recommendation method based on comprehensive feature representation and improved wide-depth model | |
| CN116993374A (en) | Model optimization method, device, equipment and medium based on deep neural network | |
| Zhang et al. | Cosine: compressive network embedding on large-scale information networks | |
| CN112597292B (en) | Question reply recommendation method, device, computer equipment and storage medium | |
| CN116109650B (en) | Point cloud instance segmentation model training method and training device | |
| CN112784008A (en) | Case similarity determining method and device, storage medium and terminal | |
| CN116467466A (en) | Knowledge graph-based code recommendation method, device, equipment and medium | |
| Chen et al. | Community Detection Based on DeepWalk Model in Large‐Scale Networks | |
| CN115240782A (en) | Drug attribute prediction method, device, electronic device and storage medium | |
| CN116415624A (en) | Model training method and device, and content recommendation method and device | |
| CN111931035B (en) | Service recommendation method, device and equipment | |
| Rizi et al. | Global and local feature learning for ego-network analysis | |
| CN118069814B (en) | Text processing method, device, electronic equipment and storage medium | |
| CN111291182A (en) | Hotspot event discovery method, device, equipment and storage medium | |
| CN117252665B (en) | Service recommendation method and device, electronic equipment and storage medium | |
| Yang et al. | Self-supervised Hypergraph Transformer with Alignment and Uniformity for Recommendation. | |
| Luan et al. | Physics-Informed Research Assistant for Theory Extraction (PIRATE) for Missing Physics Discovery | |
| Sridhar et al. | Extending Deep Neural Categorisation Models for Recommendations by Applying Gradient Based Learning | |
| CN115858942B (en) | User input-oriented serialization recommendation method and device |
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 |