CN114119142A - Information recommendation method, device and system - Google Patents

Information recommendation method, device and system Download PDF

Info

Publication number
CN114119142A
CN114119142A CN202111331942.4A CN202111331942A CN114119142A CN 114119142 A CN114119142 A CN 114119142A CN 202111331942 A CN202111331942 A CN 202111331942A CN 114119142 A CN114119142 A CN 114119142A
Authority
CN
China
Prior art keywords
product
image
clustering
products
cluster
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
Application number
CN202111331942.4A
Other languages
Chinese (zh)
Inventor
石静雯
杨勇
李征
王冬月
丁卓冶
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Jingdong Century Trading Co Ltd
Beijing Wodong Tianjun Information Technology Co Ltd
Original Assignee
Beijing Jingdong Century Trading Co Ltd
Beijing Wodong Tianjun Information Technology Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beijing Jingdong Century Trading Co Ltd, Beijing Wodong Tianjun Information Technology Co Ltd filed Critical Beijing Jingdong Century Trading Co Ltd
Priority to CN202111331942.4A priority Critical patent/CN114119142A/en
Publication of CN114119142A publication Critical patent/CN114119142A/en
Priority to PCT/CN2022/127554 priority patent/WO2023082993A1/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0631—Item recommendations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0641—Shopping interfaces
    • G06Q30/0643—Graphical representation of items or shoppers

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Business, Economics & Management (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • Probability & Statistics with Applications (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The disclosure provides an information recommendation method, device and system, and relates to the technical field of computers. The method comprises the following steps: acquiring an image of a product; determining a characterization vector for an image of a product; determining a clustering mark corresponding to the image of each product according to the characterization vector of the image of each product; according to the products and the cluster identifications corresponding to the images of the products, formulating limiting conditions for displaying different products corresponding to the same cluster identification at the same time to form a scattering rule; and utilizing the scattering rule to scatter the product recommendation result. Therefore, the scattering effect is improved, and the problem of information bundling recommendation is better solved.

Description

Information recommendation method, device and system
Technical Field
The present disclosure relates to the field of computer technologies, and in particular, to an information recommendation method, apparatus, and system.
Background
The business system may recommend some information to the user, for example, the e-commerce system may recommend some product information to the user. However, the recommendation system may recommend a bundle of similar products to the user, affecting the user experience.
Some related technologies make a breaking rule by using category information or text information of a product, and break a product recommendation result to solve the problem of information bundling recommendation.
Disclosure of Invention
Research shows that related technologies use category information or text information of products to formulate scattering rules and scatter product recommendation results, and problems that similar commodities under different categories/text information in the recommendation results cannot be scattered or dissimilar commodities under the same categories/text information are forcibly scattered and the like exist, scattering effects are affected, the problem of information bundling recommendation is not facilitated, and user experience is affected.
The cluster identification corresponding to the image of the product is determined, the limiting conditions of displaying different products corresponding to the same cluster identification at the same time are formulated according to the cluster identification, the scattering rule is formed, the product recommendation result is scattered, the scattering effect is improved, the problem of information bunching recommendation is better solved, and the user experience is improved.
Some embodiments of the present disclosure provide an information recommendation method, including:
acquiring an image of a product;
determining a characterization vector for an image of a product;
determining a clustering mark corresponding to the image of each product according to the characterization vector of the image of each product;
according to the products and the cluster identifications corresponding to the images of the products, formulating limiting conditions for displaying different products corresponding to the same cluster identification at the same time to form a scattering rule;
and utilizing the scattering rule to scatter the product recommendation result.
In some embodiments, determining, according to the characterization vector of the image of each product, a cluster identifier corresponding to the image of the product includes:
dividing the products into a plurality of initial categories according to the category information and the product words of the products;
determining a corresponding characterization vector of each initial class according to the characterization vectors of the images of the products under each initial class;
clustering the corresponding characterization vectors of each initial category to form a plurality of first-level clusters;
clustering the characterization vectors of the images of the products under each primary cluster to form a plurality of secondary clusters under each primary cluster;
and determining the primary clustering identification and the secondary clustering identification corresponding to the image of the product according to the primary clustering and the secondary clustering to which the image of the product belongs.
In some embodiments, the cluster identifier corresponding to the image of the product includes a first-level cluster identifier and a second-level cluster identifier corresponding to the image of the product; according to the clustering marks corresponding to the products and the images of the products, formulating the limiting conditions that different products corresponding to the same clustering marks are displayed simultaneously, and forming the scattering rules comprises: and in the results of limiting the adjacent first quantity in the product recommendation results, the different products corresponding to the same first-level cluster identifier display a second quantity at most, and the different products corresponding to the same second-level cluster identifier display a third quantity at most, wherein the first quantity is greater than the second quantity, and the second quantity is greater than the third quantity.
In some embodiments, determining a characterization vector for an image of a product comprises:
inputting the image of the product into the trained image vector extraction network, outputting the representation vector of the image of the product,
the image vector extraction network is obtained by training a convolutional neural network by using an image of a product and a product word of the product serving as a label, the image of the product for training is preprocessed, and the preprocessing comprises one or more of the following steps:
if the image of the product for training is a transparent image, converting the transparent image into a white background image;
randomly and horizontally turning the image of the product for training at a preset probability;
randomly rotating the image of the product for training by an angle within a preset angle according to a preset probability;
the colors of the images of the product for training are randomly shifted with a preset probability.
In some embodiments, determining, according to the characterization vector of the image of each product, the cluster identifier corresponding to the image of the product further includes:
determining the initial category of the new product according to the category information and the product words of the new product;
determining a primary cluster identifier of the image of the new product corresponding to the initial category of the new product according to the mapping relation between the initial category and the primary cluster;
and according to the distance between the characterization vector of the image of the new product and each secondary cluster under the primary cluster of the new product, taking the secondary cluster mark corresponding to the shortest distance as the secondary cluster mark of the image of the new product.
In some embodiments, the characterization vectors of the images of the products under each initial category are averaged and pooled to obtain the corresponding characterization vector of the initial category; the clustering method for forming the first-level clustering comprises k-means clustering and hierarchical clustering; the clustering method for forming the secondary clustering comprises CBSCAN clustering and canty clustering.
Some embodiments of the present disclosure provide an information recommendation method, including:
obtaining a product recommendation result;
acquiring a preset scattering rule, wherein the scattering rule comprises a limiting condition which is formulated according to the products and the cluster identifiers corresponding to the images of the products and corresponds to the same cluster identifiers and is used for displaying different products simultaneously;
and utilizing the scattering rule to scatter the product recommendation result.
In some embodiments, the method of forming the break-up rule includes:
acquiring an image of a product;
determining a characterization vector for an image of a product;
determining a clustering mark corresponding to the image of each product according to the characterization vector of the image of each product;
according to the clustering marks corresponding to the products and the images of the products, formulating the limiting conditions for displaying different products corresponding to the same clustering mark at the same time to form a scattering rule.
In some embodiments, determining, according to the characterization vector of the image of each product, a cluster identifier corresponding to the image of the product includes:
dividing the products into a plurality of initial categories according to the category information and the product words of the products;
determining a corresponding characterization vector of each initial class according to the characterization vectors of the images of the products under each initial class;
clustering the corresponding characterization vectors of each initial category to form a plurality of first-level clusters;
clustering the characterization vectors of the images of the products under each primary cluster to form a plurality of secondary clusters under each primary cluster;
and determining the primary clustering identification and the secondary clustering identification corresponding to the image of the product according to the primary clustering and the secondary clustering to which the image of the product belongs.
In some embodiments, the cluster identifier corresponding to the image of the product includes a first-level cluster identifier and a second-level cluster identifier corresponding to the image of the product; according to the clustering marks corresponding to the products and the images of the products, formulating the limiting conditions that different products corresponding to the same clustering marks are displayed simultaneously, and forming the scattering rules comprises: and in the results of limiting the adjacent first quantity in the product recommendation results, the different products corresponding to the same first-level cluster identifier display a second quantity at most, and the different products corresponding to the same second-level cluster identifier display a third quantity at most, wherein the first quantity is greater than the second quantity, and the second quantity is greater than the third quantity.
Some embodiments of the present disclosure provide an information recommendation apparatus, including:
an image acquisition module configured to acquire an image of a product;
a vector determination module configured to determine a characterization vector for an image of a product;
the cluster identifier determining module is configured to determine cluster identifiers corresponding to the images of the products according to the characterization vectors of the images of the products;
the breaking rule making module is configured to make limiting conditions for simultaneously displaying different products corresponding to the same clustering mark according to the products and the clustering marks corresponding to the images of the products to form a breaking rule;
and the scattering processing module is configured to scatter the product recommendation result by using the scattering rule.
Some embodiments of the present disclosure provide an information recommendation apparatus, including:
a recommendation result obtaining module configured to obtain a product recommendation result;
the system comprises a scattering rule acquisition module, a scattering rule display module and a scattering rule display module, wherein the scattering rule acquisition module is configured to acquire a preset scattering rule, and the scattering rule comprises a limiting condition that different products corresponding to the same clustering mark are displayed at the same time according to the products and the clustering mark corresponding to the image of the products;
and the scattering processing module is configured to scatter the product recommendation result by using the scattering rule.
Some embodiments of the present disclosure provide an information recommendation apparatus, including: a memory; and a processor coupled to the memory, the processor configured to perform the information recommendation methods of the various embodiments based on instructions stored in the memory.
Some embodiments of the present disclosure provide an information recommendation system, which includes:
a first information recommendation unit configured to form an initial product recommendation result;
and the second information recommendation unit is configured to scatter the product recommendation result by executing the information recommendation methods of the various embodiments.
Some embodiments of the present disclosure provide a non-transitory computer-readable storage medium having stored thereon a computer program that, when executed by a processor, implements the steps of the information recommendation method of the various embodiments.
Drawings
The drawings that will be used in the description of the embodiments or the related art will be briefly described below. The present disclosure can be understood more clearly from the following detailed description, which proceeds with reference to the accompanying drawings.
It is to be understood that the drawings in the following description are merely exemplary of the disclosure, and that other drawings may be derived from those drawings by one of ordinary skill in the art without undue inventive faculty.
Fig. 1 shows a flow diagram of an information recommendation method of some embodiments of the present disclosure.
Fig. 2 illustrates a schematic diagram of a process for generating image characterization vectors according to some embodiments of the present disclosure.
Fig. 3 illustrates a coarse-grained first-level clustering diagram of some embodiments of the present disclosure.
Fig. 4 shows a fine-grained secondary clustering diagram of some embodiments of the present disclosure.
Fig. 5 is a schematic structural diagram of an information recommendation device according to some embodiments of the present disclosure.
Fig. 6 shows a schematic structural diagram of an information recommendation device according to some embodiments of the present disclosure.
Fig. 7 is a schematic structural diagram of an information recommendation device according to some embodiments of the present disclosure.
Fig. 8 shows a schematic structural diagram of an information recommendation system according to some embodiments of the present disclosure.
Detailed Description
The technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the drawings in the embodiments of the present disclosure.
Unless otherwise specified, "first", "second", and the like in the present disclosure are described to distinguish different objects, and are not intended to mean size, timing, or the like.
Fig. 1 shows a flow diagram of an information recommendation method of some embodiments of the present disclosure.
As shown in fig. 1, the information recommendation method of this embodiment includes:
at step 110, an image of the product is acquired.
The image of the product is obtained according to a URL (Uniform Resource Locator) of the image of the product.
At step 120, a characterization vector for the image of the product is determined using computer vision (computer vision) techniques. The generation process of the image characterization vector is shown in fig. 2.
Determining a characterization vector for an image of a product comprises: and inputting the image of the product into the trained image vector extraction network, and outputting the representation vector of the image of the product.
The image vector extraction network is obtained by training a convolutional neural network by using an image of a product and a product word of the product serving as a label.
The above training data can be obtained by the following method. And selecting images of products with higher recent clicks as training data, wherein each image takes text information of the product as a label, and for example, a product word of the available product as a label. And filtering out product words at least containing 500 product images, wherein each product word takes 500 images, for example, and finally obtaining 700 pieces of image data. The training data may be divided into a training set, a validation set, and a test set.
The network is trained by a method of transfer learning. Firstly, obtaining a convolution neural network after initial training, wherein the initial training can be training by utilizing a large number of natural images (such as Imagenet data), and training the pre-training weight of the convolution neural network, so that the convolution neural network after the initial training has the learning capability of the bottom visual characteristics. Then, the preliminarily trained convolutional neural network is transferred to a specific information recommendation task, the convolutional neural network is further trained by the image of the product and the product word of the product serving as the label to obtain an image vector extraction network, and the weight of the convolutional neural network is finely adjusted, so that the trained network is suitable for the specific information recommendation task. The convolutional neural network uses, for example, a network such as resnet, vgg as a backbone (backbone) network. The pre-training weight is loaded, so that the learning of the network is accelerated, and the training time is shortened.
During training, the image data may be pre-processed before being input into the neural network. The pre-treatment comprises one or more of the following. 1) If the input image is a transparent image, converting the transparent image of the 4 channels into a white background image of the 3 channels, wherein the 3 channels are RGB channels, and the 4 channels have one more transparency channel than the 3 channels; 2) randomly intercepting a part of the image, wherein the area ratio of an intercepted area is between [0.7 and 1.0], and the aspect ratio is between [0.75 and 1.33 ]; 3) image resizing to 224 × 224 size; 4) randomly and horizontally turning over according to a preset probability such as 0.5; 5) randomly rotating a preset angle such as an angle within 45 degrees according to a preset probability such as 0.5; 6) randomly shifting image colors with a preset probability such as 0.5; 7) pixel values are normalized to between [1, -1 ].
By the preprocessing of 1)4)5)6) above, the network is made to concentrate on the characteristics of the product itself, and is insensitive to the background, angle, color, etc. of the product image. The image is standardized in size or pixels by the preprocessing of 2)3)7) to improve the training effect.
In the training process, in order to reduce the operation amount of subsequent operation, the vector generated by the backbone network is further reduced in dimension. Taking the backbone network as resnet101 as an example, after the image passes through the convolutional layer of resnet101 and an average pooling (averaging) operation, K is obtained1Dimension feature vector, and then three layers of dimension reduction networks are accessed in the back, which are respectively from K1Sequentially reducing dimension to K2、K3、K4Dimension, K1＞K2＞K3＞K4. In order to reduce information loss in the dimension reduction process, the dimension reduction network adopts a prelu activation function. Finally, obtaining a logistic regression value (logits) corresponding to each product word through one layer of linear transformation, and changing the logistic regression value (logits) into [0, 1] through a sigmoid function]Probability values of the ranges. The method adopts a multi-label classification objective function, one image can be predicted into a plurality of product words which are independent of each other, each class is equivalent to a two-classification task, and the two-classification b is usedintake-cross-entry as a function of loss. Training uses adam optimizer for back propagation learning, setting an initial learning rate of 0.001 and dynamic adjustment of learning rate, and single-machine multi-card acceleration with horosod. And selecting the model with the highest top-1, top-5 and top-10 recall rates (recall) of the product words in the test set as the model of the prediction stage.
In step 130, according to the characterization vector of the image of each product, a cluster identifier corresponding to the image of the product is determined.
If the number of the products is small, a clustering algorithm can be adopted to perform primary clustering on the characterization vectors of the images of the products, and the clustering identification corresponding to the images of the products is directly determined.
If the number of the products is large, twice clustering of coarse granularity and fine granularity can be carried out, and the first-level clustering identification and the second-level clustering identification corresponding to the images of the products are determined. Thus, the amount of calculation is reduced. The latter case is described below with emphasis.
According to the characterization vectors of the images of the products, the cluster identifier corresponding to the image of the existing product is determined through clustering, which is detailed in step 130.1, and the cluster identifier corresponding to the image of the new product is determined through matching, which is detailed in step 130.2.
Step 130.1 for existing products, determining a cluster identifier corresponding to the image of the existing product by clustering according to the characterization vector of the image of each existing product, and including: dividing the existing products into a plurality of initial categories according to the category information and the product words of the products; determining a corresponding characterization vector of each initial class according to the characterization vectors of the images of the existing products under each initial class; clustering the corresponding characterization vectors of each initial category to form a plurality of first-level clusters, also called coarse-grained clusters, as shown in fig. 3; clustering the characterization vectors of the images of the existing products under each primary cluster to form a plurality of secondary clusters under each primary cluster, which are also called fine-grained clusters, as shown in FIG. 4; and determining the primary clustering identification and the secondary clustering identification corresponding to the image of the existing product according to the primary clustering and the secondary clustering to which the image of the existing product belongs. Thus determining coarse-grained clustering and fine-grained clustering corresponding to the images.
For example, the product is divided into N C1_ PW categories (namely initial categories) according to the dimension of the primary Product Word (PW) spliced by the primary category (C1); according to the requirement, C1_ PW with the product quantity less than 100 under the C1_ PW can be filtered, data noise caused by wrong product words is eliminated, and if the remaining N C1_ PW are assumed, N is equal to or less than N; randomly and uniformly sampling the products under the remaining n C1_ PWs, wherein each C1_ PW is sampled by 1 ten thousand products for example, and each product corresponds to one K4Dimensional characterization vector (image emb), K for all sampled products at each C1_ PW4The characterization vector of dimension is subjected to average pooling (mean pooling) operation to obtain K characterizing each C1_ PW4Dimension vector (C1_ PW emb), K of n total C1_ PW4A dimension vector; k for n C1_ PW4The dimension vectors are clustered to obtain m coarse-grained categories (or called coarse-grained VIDs, i.e. first-level clustering). Available clustering algorithms are k-means clustering, Hierarchical clustering, etc. Taking hierarchical clustering as an example, the samples to be clustered are K of n C1_ PW4Dimension vector, 1) initializing, and regarding each sample as a cluster; 2) calculating the similarity among the clusters; 3) searching two clusters which are the most similar according to the similarity, and classifying the two clusters into one class; 4) and repeating the steps 2) and 3) until the similarity between each class is larger than a preset similarity threshold, and the rest classes are final clustering results. And obtaining the coarse-granularity VID attribute of each product according to the C1_ PW corresponding to the product and the mapping relation between the C1_ PW and the coarse-granularity VID. And then, quickly clustering the corresponding product image vector under each coarse-granularity VID to obtain the fine-granularity VID under each coarse-granularity VID (namely secondary clustering). Available fast clustering methods suitable for large data sets include CBSCAN, canopy, etc. Taking canty as an example, the samples to be clustered are all product image vectors at each coarse-grained VID. 1) Setting an initial distance threshold value as T when the sample set to be clustered is S1、T2And T is1>T2(ii) a 2) Stacking and selecting a sample A in the S, and calculating the distance d between the A and other samples in the set S by using a rough distance calculation mode such as Euclidean distance; 3) according to the distance d in 2), d is less than T1OfThis data vector is divided into a canty, with d smaller than T2Removing the sample data vector of (a) from the candidate center vector S; 4) and repeating the step 2) and the step 3) until the candidate center vector S is empty, finishing clustering and enabling each sphere to be a fine-grained VID.
Step 130.2 for the new product, determining the cluster identifier corresponding to the image of the new product by matching according to the characterization vector of the image of each product, including: determining the initial category of the new product according to the category information and the product words of the new product; determining a primary cluster identifier of the image of the new product corresponding to the initial category of the new product according to the mapping relation between the initial category and the primary cluster; and according to the distance between the characterization vector of the image of the new product and each secondary cluster under the primary cluster of the new product, taking the secondary cluster mark corresponding to the shortest distance as the secondary cluster mark of the image of the new product.
For example, for a new product, the new product C1_ PW category is obtained according to the class C1 of the product and the text information (such as the main product word PW), and the corresponding coarse-grained VID of the new product is determined according to the mapping relation between the C1_ PW and the coarse-grained VID. And then, extracting the characterization vector of the new product image, calculating the distance between the characterization vector of the new product and the fine-grained VID under the coarse-grained VID, and determining the corresponding fine-grained VID with the closest distance as the fine-grained VID of the new product.
In step 140, according to the clustering marks corresponding to the products and the images of the products, the limiting conditions for displaying different products corresponding to the same clustering mark at the same time are formulated, and a breaking rule is formed.
If the cluster identifier corresponding to the image of the product comprises a first-level cluster identifier and a second-level cluster identifier corresponding to the image of the product; the breaking rules formed include: in the result of limiting the adjacent first number (e.g. 4) in the product recommendation result, the different products corresponding to the same first-level cluster identifier exhibit at most a second number (e.g. 2), and the different products corresponding to the same second-level cluster identifier exhibit at most a third number (e.g. 1), wherein the first number is greater than the second number, and the second number is greater than the third number.
If the image of the product corresponds to a cluster identifier, the formed scattering rule comprises the following steps: and limiting a fourth number (for example, 5) of adjacent results in the product recommendation result, wherein the fourth number is greater than the fifth number, and the different products corresponding to the same cluster identifier exhibit at most a fifth number (for example, 1).
In step 150, the product recommendation result is broken up according to the breaking-up rule.
Obtaining a product recommendation result; acquiring a preset scattering rule, wherein the scattering rule comprises a limiting condition which is formulated according to the products and the cluster identifiers corresponding to the images of the products and corresponds to the same cluster identifiers and is used for displaying different products simultaneously; and utilizing the scattering rule to scatter the product recommendation result.
The initial product recommendation result may be formed according to a search keyword input by a user or according to other preset keywords, and the disclosure does not limit how to form the initial product recommendation result.
Taking the first breaking rule in step 140 as an example, in the results of the adjacent first number (e.g., 4) in the product recommendation results, if the number of different products corresponding to the same primary cluster identifier is greater than the second number (e.g., 2), the second number (e.g., 2) is displayed at most, and the products exceeding the second number (e.g., 2) can be displayed in the recommendation results of the next first number for judgment, similarly, if the number of different products corresponding to the same secondary cluster identifier is greater than the third number (e.g., 1), the third number (e.g., 1) is displayed at most, and the products exceeding the third number (e.g., 1) can be displayed in the recommendation results of the next first number for judgment.
According to the embodiment, the cluster identifier corresponding to the image of the product is determined, the limiting conditions for displaying different products corresponding to the same cluster identifier at the same time are formulated accordingly, the scattering rule is formed, the product recommendation result is scattered, the scattering effect is improved, and the problem of information bunching recommendation is better solved. In addition, if the scattering rule formed by the thick-thin two-level clustering marks is used for scattering, the calculation amount can be reduced.
Fig. 5 is a schematic structural diagram of an information recommendation device according to some embodiments of the present disclosure.
As shown in fig. 5, the information recommendation apparatus 500 of this embodiment includes the following modules.
An image acquisition module 510 configured to acquire an image of a product;
a vector determination module 520 configured to determine a characterization vector for an image of a product;
a cluster identifier determining module 530 configured to determine a cluster identifier corresponding to the image of the product according to the characterization vector of the image of each product;
a breaking rule making module 540 configured to make a limiting condition that different products corresponding to the same cluster identifier are displayed simultaneously according to the product and the cluster identifier corresponding to the image of the product, so as to form a breaking rule;
and a breaking-up processing module 550 configured to break up the product recommendation result by using the breaking-up rule.
In some embodiments, the cluster identifier determining module 530 configured to determine, according to the characterization vector of the image of each product, the cluster identifier corresponding to the image of the product includes:
dividing the products into a plurality of initial categories according to the category information and the product words of the products;
determining a corresponding characterization vector of each initial class according to the characterization vectors of the images of the products under each initial class;
clustering the corresponding characterization vectors of each initial category to form a plurality of first-level clusters;
clustering the characterization vectors of the images of the products under each primary cluster to form a plurality of secondary clusters under each primary cluster;
and determining the primary clustering identification and the secondary clustering identification corresponding to the image of the product according to the primary clustering and the secondary clustering to which the image of the product belongs.
The cluster identification corresponding to the image of the product comprises a first-level cluster identification and a second-level cluster identification corresponding to the image of the product.
In some embodiments, the breaking rule making module 540 is configured to limit the number of the adjacent first number of the product recommendation results, wherein different products corresponding to the same first-level cluster identifier exhibit at most a second number, and different products corresponding to the same second-level cluster identifier exhibit at most a third number, wherein the first number is greater than the second number, and the second number is greater than the third number.
In some embodiments, the vector determination module 520 is configured to input the image of the product into a trained image vector extraction network, and output a characterization vector of the image of the product, where the image vector extraction network is obtained by training a convolutional neural network using the image of the product and a product word of the product as a tag, and the image of the product for training is preprocessed, where the preprocessing includes one or more of the following:
if the image of the product for training is a transparent image, converting the transparent image into a white background image;
randomly and horizontally turning the image of the product for training at a preset probability;
randomly rotating the image of the product for training by an angle within a preset angle according to a preset probability;
the colors of the images of the product for training are randomly shifted with a preset probability.
In some embodiments, the cluster identification determination module 530 is configured to determine an initial category of a new product according to the category information and the product words of the new product;
determining a primary cluster identifier of the image of the new product corresponding to the initial category of the new product according to the mapping relation between the initial category and the primary cluster;
and according to the distance between the characterization vector of the image of the new product and each secondary cluster under the primary cluster of the new product, taking the secondary cluster mark corresponding to the shortest distance as the secondary cluster mark of the image of the new product.
In some embodiments, the characterization vectors of the images of the respective products under each initial category are averaged and pooled to obtain the corresponding characterization vector for that initial category.
In some embodiments, the clustering method for forming the first-level cluster comprises k-means clustering and hierarchical clustering.
In some embodiments, the clustering method for forming the secondary clusters includes CBSCAN clustering and sphere clustering.
Fig. 6 shows a schematic structural diagram of an information recommendation device according to some embodiments of the present disclosure.
As shown in fig. 6, the information recommendation apparatus 600 of this embodiment includes the following modules.
A recommendation obtaining module 610 configured to obtain product recommendations.
A breaking rule obtaining module 620 configured to obtain a preset breaking rule, where the breaking rule includes a restriction condition that different products corresponding to the same cluster identifier are displayed at the same time, the restriction condition being formulated according to the product and the cluster identifier corresponding to the image of the product; the forming method of the break-up rule is described in the foregoing, and is not described in detail here.
And the scattering processing module 630 is configured to scatter the product recommendation result by using the scattering rule.
Fig. 7 is a schematic structural diagram of an information recommendation device according to some embodiments of the present disclosure.
As shown in fig. 7, the information recommendation apparatus 700 of this embodiment includes: a memory 710 and a processor 720 coupled to the memory 710, the processor 720 configured to execute the information recommendation method in any of the embodiments based on instructions stored in the memory 710.
The information recommendation apparatus 700 may be, for example, a second information recommendation unit.
Memory 710 may include, for example, system memory, fixed non-volatile storage media, and the like. The system memory stores, for example, an operating system, an application program, a Boot Loader (Boot Loader), and other programs.
Processor 720 may be implemented, for example, as discrete hardware components, such as a general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic device, discrete gates, or transistors.
The apparatus 700 may also include an input-output interface 730, a network interface 740, a storage interface 750, and the like. These interfaces 730, 740, 750, as well as the memory 710 and the processor 720, may be connected, for example, by a bus 760. The input/output interface 730 provides a connection interface for input/output devices such as a display, a mouse, a keyboard, and a touch screen. The network interface 740 provides a connection interface for various networking devices. The storage interface 750 provides a connection interface for external storage devices such as an SD card and a usb disk. Bus 760 may employ any of a variety of bus architectures. For example, bus structures include, but are not limited to, Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MCA) bus, and Peripheral Component Interconnect (PCI) bus.
FIG. 8 illustrates a schematic diagram of an information recommendation system of some embodiments of the present disclosure.
As shown in fig. 8, the information recommendation system 800 of this embodiment includes:
a first information recommending unit 810 configured to form an initial product recommendation result;
and a second information recommendation unit 820 configured to scatter the product recommendation result by performing the information recommendation method of each embodiment.
The first information recommending unit 810 may form an initial product recommendation result according to a search keyword input by a user, or form an initial product recommendation result according to other preset keywords. The present disclosure does not define how the initial product recommendations are formed.
Some embodiments of the present disclosure provide a non-transitory computer-readable storage medium having stored thereon a computer program that, when executed by a processor, implements the steps of the information recommendation method of the embodiments.
As will be appreciated by one skilled in the art, embodiments of the present disclosure may be provided as a method, system, or computer program product. Accordingly, the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product embodied on one or more non-transitory computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and so forth) having computer program code embodied therein.
The present disclosure is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the disclosure. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The above description is only exemplary of the present disclosure and is not intended to limit the present disclosure, so that any modification, equivalent replacement, or improvement made within the spirit and principle of the present disclosure should be included in the scope of the present disclosure.

Claims (15)

1. An information recommendation method, comprising:
acquiring an image of a product;
determining a characterization vector for an image of a product;
determining a clustering mark corresponding to the image of each product according to the characterization vector of the image of each product;
according to the products and the cluster identifications corresponding to the images of the products, formulating limiting conditions for displaying different products corresponding to the same cluster identification at the same time to form a scattering rule;
and utilizing the scattering rule to scatter the product recommendation result.
2. The method of claim 1, wherein determining the cluster identifier corresponding to the image of the product according to the characterization vector of the image of each product comprises:
dividing the products into a plurality of initial categories according to the category information and the product words of the products;
determining a corresponding characterization vector of each initial class according to the characterization vectors of the images of the products under each initial class;
clustering the corresponding characterization vectors of each initial category to form a plurality of first-level clusters;
clustering the characterization vectors of the images of the products under each primary cluster to form a plurality of secondary clusters under each primary cluster;
and determining the primary clustering identification and the secondary clustering identification corresponding to the image of the product according to the primary clustering and the secondary clustering to which the image of the product belongs.
3. The method of claim 1, wherein the cluster identifier corresponding to the image of the product comprises a first-level cluster identifier and a second-level cluster identifier corresponding to the image of the product;
according to the clustering marks corresponding to the products and the images of the products, formulating the limiting conditions that different products corresponding to the same clustering marks are displayed simultaneously, and forming the scattering rules comprises:
and in the results of limiting the adjacent first quantity in the product recommendation results, the different products corresponding to the same first-level cluster identifier display a second quantity at most, and the different products corresponding to the same second-level cluster identifier display a third quantity at most, wherein the first quantity is greater than the second quantity, and the second quantity is greater than the third quantity.
4. The method of claim 1, wherein determining a characterization vector for an image of a product comprises:
inputting the image of the product into the trained image vector extraction network, outputting the representation vector of the image of the product,
the image vector extraction network is obtained by training a convolutional neural network by using an image of a product and a product word of the product serving as a label, the image of the product for training is preprocessed, and the preprocessing comprises one or more of the following steps:
if the image of the product for training is a transparent image, converting the transparent image into a white background image;
randomly and horizontally turning the image of the product for training at a preset probability;
randomly rotating the image of the product for training by an angle within a preset angle according to a preset probability;
the colors of the images of the product for training are randomly shifted with a preset probability.
5. The method of claim 2, wherein determining the cluster identifier corresponding to the image of the product according to the characterization vector of the image of each product further comprises:
determining the initial category of the new product according to the category information and the product words of the new product;
determining a primary cluster identifier of the image of the new product corresponding to the initial category of the new product according to the mapping relation between the initial category and the primary cluster;
and according to the distance between the characterization vector of the image of the new product and each secondary cluster under the primary cluster of the new product, taking the secondary cluster mark corresponding to the shortest distance as the secondary cluster mark of the image of the new product.
6. The method of claim 2,
carrying out average pooling on the characterization vectors of the images of the products under each initial category to obtain the corresponding characterization vectors of the initial categories;
the clustering method for forming the first-level clustering comprises k-means clustering and hierarchical clustering;
the clustering method for forming the secondary clustering comprises CBSCAN clustering and canty clustering.
7. An information recommendation method, comprising:
obtaining a product recommendation result;
acquiring a preset scattering rule, wherein the scattering rule comprises a limiting condition which is formulated according to the products and the cluster identifiers corresponding to the images of the products and corresponds to the same cluster identifiers and is used for displaying different products simultaneously;
and utilizing the scattering rule to scatter the product recommendation result.
8. The method of claim 7, wherein the method of forming the break-up rule comprises:
acquiring an image of a product;
determining a characterization vector for an image of a product;
determining a clustering mark corresponding to the image of each product according to the characterization vector of the image of each product;
according to the clustering marks corresponding to the products and the images of the products, formulating the limiting conditions for displaying different products corresponding to the same clustering mark at the same time to form a scattering rule.
9. The method of claim 8, wherein determining the cluster identifier corresponding to the image of the product according to the characterization vector of the image of each product comprises:
dividing the products into a plurality of initial categories according to the category information and the product words of the products;
determining a corresponding characterization vector of each initial class according to the characterization vectors of the images of the products under each initial class;
clustering the corresponding characterization vectors of each initial category to form a plurality of first-level clusters;
clustering the characterization vectors of the images of the products under each primary cluster to form a plurality of secondary clusters under each primary cluster;
and determining the primary clustering identification and the secondary clustering identification corresponding to the image of the product according to the primary clustering and the secondary clustering to which the image of the product belongs.
10. The method of claim 8, wherein the cluster identifier corresponding to the image of the product comprises a first-level cluster identifier and a second-level cluster identifier corresponding to the image of the product;
according to the clustering marks corresponding to the products and the images of the products, formulating the limiting conditions that different products corresponding to the same clustering marks are displayed simultaneously, and forming the scattering rules comprises:
and in the results of limiting the adjacent first quantity in the product recommendation results, the different products corresponding to the same first-level cluster identifier display a second quantity at most, and the different products corresponding to the same second-level cluster identifier display a third quantity at most, wherein the first quantity is greater than the second quantity, and the second quantity is greater than the third quantity.
11. An information recommendation apparatus, comprising:
an image acquisition module configured to acquire an image of a product;
a vector determination module configured to determine a characterization vector for an image of a product;
the cluster identifier determining module is configured to determine cluster identifiers corresponding to the images of the products according to the characterization vectors of the images of the products;
the breaking rule making module is configured to make limiting conditions for simultaneously displaying different products corresponding to the same clustering mark according to the products and the clustering marks corresponding to the images of the products to form a breaking rule;
and the scattering processing module is configured to scatter the product recommendation result by using the scattering rule.
12. An information recommendation apparatus, comprising:
a recommendation result obtaining module configured to obtain a product recommendation result;
the system comprises a scattering rule acquisition module, a scattering rule display module and a scattering rule display module, wherein the scattering rule acquisition module is configured to acquire a preset scattering rule, and the scattering rule comprises a limiting condition that different products corresponding to the same clustering mark are displayed at the same time according to the products and the clustering mark corresponding to the image of the products;
and the scattering processing module is configured to scatter the product recommendation result by using the scattering rule.
13. An information recommendation apparatus, comprising: a memory; and a processor coupled to the memory, the processor configured to perform the information recommendation method of any of claims 1-10 based on instructions stored in the memory.
14. An information recommendation system, comprising:
a first information recommendation unit configured to form an initial product recommendation result;
a second information recommendation unit configured to scatter the product recommendation result by performing the information recommendation method of any one of claims 1 to 10.
15. A non-transitory computer readable storage medium, having stored thereon a computer program which, when executed by a processor, carries out the steps of the information recommendation method of any one of claims 1-10.
CN202111331942.4A 2021-11-11 2021-11-11 Information recommendation method, device and system Pending CN114119142A (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202111331942.4A CN114119142A (en) 2021-11-11 2021-11-11 Information recommendation method, device and system
PCT/CN2022/127554 WO2023082993A1 (en) 2021-11-11 2022-10-26 Information recommendation method, apparatus and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111331942.4A CN114119142A (en) 2021-11-11 2021-11-11 Information recommendation method, device and system

Publications (1)

Publication Number Publication Date
CN114119142A true CN114119142A (en) 2022-03-01

Family

ID=80378355

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111331942.4A Pending CN114119142A (en) 2021-11-11 2021-11-11 Information recommendation method, device and system

Country Status (2)

Country Link
CN (1) CN114119142A (en)
WO (1) WO2023082993A1 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2023082993A1 (en) * 2021-11-11 2023-05-19 北京沃东天骏信息技术有限公司 Information recommendation method, apparatus and system

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5445339B2 (en) * 2010-06-08 2014-03-19 ソニー株式会社 Content recommendation device and content recommendation method
CN111191112A (en) * 2019-07-12 2020-05-22 腾讯科技（深圳）有限公司 Electronic reading data processing method, device and storage medium
CN112565903B (en) * 2019-09-10 2022-11-22 北京达佳互联信息技术有限公司 Video recommendation method and device, server and storage medium
CN111125460B (en) * 2019-12-24 2022-02-25 腾讯科技（深圳）有限公司 Information recommendation method and device
CN111160254A (en) * 2019-12-30 2020-05-15 杭州趣维科技有限公司 Classification method and system suitable for short video recommendation scattering
CN113688167A (en) * 2021-01-15 2021-11-23 稿定（厦门）科技有限公司 Deep interest capture model construction method and device based on deep interest network
CN114119142A (en) * 2021-11-11 2022-03-01 北京沃东天骏信息技术有限公司 Information recommendation method, device and system

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2023082993A1 (en) * 2021-11-11 2023-05-19 北京沃东天骏信息技术有限公司 Information recommendation method, apparatus and system

Also Published As

Publication number Publication date
WO2023082993A1 (en) 2023-05-19

Similar Documents

Publication Publication Date Title
CN106651057B (en) Mobile terminal user age prediction method based on installation package sequence list
CN112434721A (en) Image classification method, system, storage medium and terminal based on small sample learning
CN107683469A (en) A kind of product classification method and device based on deep learning
Wu et al. Regularized densely-connected pyramid network for salient instance segmentation
CN109919252B (en) Method for generating classifier by using few labeled images
CN109213862B (en) Object recognition method and device, and computer-readable storage medium
CN111931859B (en) Multi-label image recognition method and device
CN110059596A (en) A kind of image-recognizing method, device, medium and electronic equipment
CN114187311A (en) Image semantic segmentation method, device, equipment and storage medium
CN110990563A (en) Artificial intelligence-based traditional culture material library construction method and system
CN110909817B (en) Distributed clustering method and system, processor, electronic device and storage medium
CN110413825B (en) Street-clapping recommendation system oriented to fashion electronic commerce
WO2023082993A1 (en) Information recommendation method, apparatus and system
CN111783812A (en) Method and device for identifying forbidden images and computer readable storage medium
CN110852351A (en) Image-based garbage classification method and device, terminal equipment and storage medium
CN112579730A (en) High-expansibility multi-label text classification method and device
Dong et al. A supervised dictionary learning and discriminative weighting model for action recognition
CN115331004A (en) Zero sample semantic segmentation method and device based on meaningful learning
CN113591881B (en) Intention recognition method and device based on model fusion, electronic equipment and medium
CN114416986A (en) Text data cleaning method and device and storage medium
CN115587297A (en) Method, apparatus, device and medium for constructing image recognition model and image recognition
Lin et al. Intelligent identification of pavement cracks based on PSA-Net
CN112633394A (en) Intelligent user label determination method, terminal equipment and storage medium
CN110532384A (en) A kind of multitask dictionary list classification method, system, device and storage medium
CN111898704B (en) Method and device for clustering content samples

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