WO2021031410A1 - Commodity information prediction model-based commodity information pushing method and apparatus, non-volatile readable storage medium, and computer device - Google Patents

Commodity information prediction model-based commodity information pushing method and apparatus, non-volatile readable storage medium, and computer device Download PDF

Info

Publication number
WO2021031410A1
WO2021031410A1 PCT/CN2019/118488 CN2019118488W WO2021031410A1 WO 2021031410 A1 WO2021031410 A1 WO 2021031410A1 CN 2019118488 W CN2019118488 W CN 2019118488W WO 2021031410 A1 WO2021031410 A1 WO 2021031410A1
Authority
WO
WIPO (PCT)
Prior art keywords
product
information
commodity
prediction model
text information
Prior art date
Application number
PCT/CN2019/118488
Other languages
French (fr)
Chinese (zh)
Inventor
邓悦
金戈
徐亮
Original Assignee
平安科技（深圳）有限公司
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 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2021031410A1 publication Critical patent/WO2021031410A1/en

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

Definitions

  • This application relates to the field of information processing technology, in particular to a commodity information push method and device based on a commodity information prediction model, a non-volatile readable storage medium, and computer equipment.
  • the existing methods for recommending products based on user information mainly include two: 1) recommending corresponding products for users based on user historical purchasing behavior information; 2) recommending corresponding products for users based on quantifiable attribute information of the products themselves. Recommending corresponding products to users based on user historical purchasing behavior information has the problem of cold start; recommending corresponding products to users based on the quantifiable attribute information of the products themselves.
  • manual input of product information is required to achieve The user's product push is not suitable for mass products, and neither of the above two methods can achieve accurate push functions based on the user's real-time needs.
  • this application provides a product information push method and device based on a product information prediction model, non-volatile readable storage media, and computer equipment.
  • the main purpose is to solve the problem that existing product information push methods cannot meet the real-time needs of users.
  • a commodity information push method based on a commodity information prediction model including:
  • a plurality of commodity information corresponding to the plurality of commodity introduction text information are respectively displayed.
  • a product information push device based on a product information prediction model including:
  • the construction module is used to construct a data candidate set according to the acquired text information of commodity requirements
  • the training module is used to train the initial product information prediction model by using the constructed data candidate set to obtain the trained product information prediction model
  • the matching module is used to input the acquired product keyword information in the product search request into the trained product information prediction model to obtain multiple product introduction text information that matches the product keyword information;
  • the display module is used to respectively display a plurality of commodity information corresponding to the plurality of commodity introduction text information.
  • a non-volatile readable storage medium having computer readable instructions stored thereon, and when the program is executed by a processor, the foregoing method for pushing commodity information based on a commodity information prediction model is realized.
  • a computer device including a non-volatile readable storage medium, a processor, and computer readable instructions stored on the non-volatile readable storage medium and running on the processor
  • the processor executes the program, the foregoing method for pushing commodity information based on the commodity information prediction model is implemented.
  • the commodity information push method and device based on the commodity information prediction model, non-volatile readable storage medium, and computer equipment provided by this application are compared with the existing technical solutions for commodity information push.
  • the keyword information is input to the trained product information prediction model to obtain multiple product introduction text information matching the product keyword information, and to display multiple product information corresponding to the multiple product introduction text information respectively. In this way, accurate product pushes can be achieved for users' real-time needs.
  • the trained product information prediction model can recommend the product information with a high degree of matching with the product demand text information according to the product demand text information input by the user.
  • FIG. 1 shows a schematic flowchart of a method for pushing commodity information based on a commodity information prediction model provided by an embodiment of the present application
  • FIG. 2 shows a schematic flowchart of another method for pushing commodity information based on a commodity information prediction model provided by an embodiment of the present application
  • Fig. 3 shows a schematic structural diagram of a product information push device based on a product information prediction model provided by an embodiment of the present application.
  • the user’s historical purchase record information is collected, and the product demand text information for different products is obtained as a data sample, and the acquired multiple product demand text information belonging to the same product is marked, so as to obtain the text information for training The data candidate set of the product information prediction model.
  • the initial product push prediction model is constructed based on a neural network model (for example, a convolutional neural network model, a recurrent neural network model, or a probabilistic neural network model), and the framework of the initial product push prediction model is specifically:
  • a neural network model for example, a convolutional neural network model, a recurrent neural network model, or a probabilistic neural network model
  • the first layer the input layer, format the input information, and the input information is the data candidate set;
  • the second layer encoding layer, which performs one-hot encoding on the input information after format processing in units of characters to obtain the corresponding semantic vector;
  • mapping layer the semantic vector is processed by multiple fully connected layers containing the activation function tanh to obtain corresponding multiple output results
  • the seventh layer the matching layer, which performs similarity calculations on multiple output results to obtain multiple similarity calculation results;
  • the eighth layer the Softmax layer, the activation function Softmax is used to normalize the calculated similarity calculation results to obtain the similarity probability.
  • the sixth layer obtains the product introduction text information used for similarity calculation, and calculates through the seventh and eighth layers according to the mapping result output by the sixth layer to obtain multiple product introduction text information matching the product keyword information.
  • multiple product information corresponding to multiple product introduction text information is acquired, and the multiple acquired product information is sorted in descending order of similarity and displayed for users to browse.
  • a data candidate set can be constructed according to the obtained product demand text information, and the constructed data candidate set can be used to train the initial product information prediction model to obtain a trained product information prediction model, so as to
  • the obtained product keyword information in the product search request is input into the trained product information prediction model to obtain multiple product introduction text information that matches the product keyword information, and the product introduction text information is respectively combined with the multiple product introduction text information.
  • Corresponding multiple product information is displayed, so as to achieve accurate product push for users' real-time needs.
  • the product information prediction model trained in this application can recommend product information that matches the product demand text information for the user according to the product demand text information input by the user.
  • FIG. Methods include:
  • the commodity demand text information includes commodity keyword information, and multiple commodity introduction text information corresponding to the commodity keyword information; wherein, it is used to construct data
  • the multiple product introduction text information corresponding to the product keyword information in the product demand text information of the candidate set includes the first product introduction text information of the product that the user has purchased and the second product introduction text information of the product that the user has not purchased.
  • the commodity demand text information includes: commodity keyword information (for example, the input commodity name, commodity name keywords), and each commodity name keyword corresponds to multiple keywords with the same commodity name or commodity name
  • the product introduction text information Mark the product introduction text information with the same product name or product name keywords, that is, mark the first product introduction text information of the product that the user has purchased as a positive sample, and mark the second product introduction text of the product that the user has not purchased It is a negative sample and serves as a data candidate set.
  • the first product introduction text information in the data candidate set is a positive sample
  • the second product introduction text information is a negative sample
  • the product keyword information is separated from the first product introduction text information and multiple second
  • the product introduction text information is calculated by distance similarity to obtain multiple similarity values.
  • the first layer (input layer) of the initial product information prediction model is used to preprocess the input information.
  • the input information is a data candidate set.
  • the number of characters of the product name keywords in the product demand text information in the data candidate set No more than 30 characters, and the number of characters in the product introduction text information with the same product name or product name keywords should not exceed 500 characters. If the number of characters is less than the required number of characters, add 0 to the end of the product name keywords and the product introduction text with the same product name or product name keywords; if the number of characters is more than the required number of characters, the Product name keywords, and the excess part of the product introduction text with the same product name or product name keywords are directly discarded.
  • the sample mark of the product introduction text information in the product demand text information is specifically that the first product introduction text information corresponding to the product purchased by the user is a positive sample, which corresponds to a product keyword information by default, and the product information purchased by the user ;
  • the multiple second product introduction text information corresponding to multiple products that the user has not purchased are multiple negative samples, which correspond to one product keyword information and multiple product information by default.
  • the second layer (coding layer) of the initial product information prediction model is used to perform one-hot encoding on the preprocessed product demand text information in character units, that is, the product name or name keyword belongs to the first sample of the positive sample.
  • One product introduction text and multiple second product introduction texts belonging to negative samples are encoded to obtain corresponding semantic vectors respectively.
  • one-hot encoding refers to the use of N-bit status registers to encode N states, each state has an independent register bit, and it is guaranteed that only one state is valid at any time.
  • the semantic vector of the word is denoted as Q
  • the semantic vector of the first product introduction text belonging to the positive sample is denoted as D 1
  • the semantic vector of multiple second product introduction texts belonging to the negative sample is denoted as D 2 -D n .
  • the obtained semantic vector is passed through multiple fully connected layers containing the activation function tanh to obtain the output result for characterizing the product keyword information , And the output result used to characterize the text information of the product introduction.
  • the input data of the third layer is the output data of the second layer (Q, D 1 , D 2 ,..., D n ), and the random initialization parameter matrix is W, b.
  • the random initialization parameter matrix and activation calculates and obtains the output result.
  • the specific calculation formula is:
  • the random initialization parameter matrix of the fourth layer is W 1 , b 1 , and the specific calculation formula is:
  • the fifth and sixth floors are the same as the third and fourth floors.
  • the obtained output results used to characterize product keyword information are respectively similar to the output results used to characterize product introduction text information. Degree calculation to obtain multiple similarity values.
  • the specific calculation formula is:
  • R(Q, D) represents the similarity value between the output result yQ used to characterize the product keyword information and the output result yD used to characterize the product introduction text information
  • cosine(y Q , y D ) represents two
  • the cosine distance of the semantic vector is the semantic similarity.
  • the activation function Softmax in the eighth layer (Softmax layer) of the initial product information prediction model is used to normalize the calculated similarity values to obtain the product keyword information Q and the first The similarity probability of the product introduction text information D 1.
  • the specific calculation formula is:
  • Q) represents the probability of similarity between the product keyword information Q and the first product introduction text information D1
  • ⁇ is the weight parameter, that is, the smoothing factor of Softmax.
  • step 204 may specifically include: determining the network parameters of the product information prediction model by performing maximum likelihood estimation on the similarity probability; and according to the determined product The network parameters of the information prediction model are used to obtain a trained commodity information prediction model.
  • the loss function is determined based on the similarity probability Q first keyword information commodity product description text message D 1, and the loss function for the log-likelihood function loss, i.e. all goods based on the first key data Q 1.
  • the maximum likelihood estimation of the similarity probability of the product introduction text information D 1 Train the initial product push prediction model to obtain the trained product push prediction model so that the product keyword information Q matches the first product introduction text information D The probability of 1 is maximized, and the loss function is minimized.
  • the specific calculation formula of the loss function is:
  • the product keyword information in the product search request and multiple product introduction text information matching the product keywords are acquired.
  • character processing is performed on the obtained product keywords and multiple product introduction text information to obtain a product keyword with a uniform number of characters and multiple product introduction text information, and use product push for the unified product keywords
  • the product keyword module in the prediction model is encoded, and the unified product introduction text information is encoded using the positive sample module in the product push prediction model to obtain the semantic vector Q used to represent the product keywords, and multiple The semantic vector D of the product introduction text information.
  • step 208 may specifically include: distance the semantic vectors corresponding to the product keyword information from the semantic vectors corresponding to the acquired multiple product introduction text information. Similarity calculation to obtain multiple similarity values; to sort the calculated similarity values in descending order to obtain a descending sorting result; according to the preset merchandise matching value, according to the descending sorting result to obtain the keywords with the merchandise Multiple product introduction text information with matching information.
  • the preset product matching value can be expressed as all the products with more than 90% similarity, or it can be expressed as the products in the preset ranking (for example, TOP100) in the descending order result. There is no matching value for the preset product here. Set standards and set dimensions for specific restrictions.
  • multiple product information corresponding to the multiple product introduction text information are obtained, and a product list is generated, that is, the product list that is recommended to the user is determined Multiple products.
  • a data candidate set is constructed based on the acquired commodity demand text information, and the constructed data candidate set is used to train the initial commodity information prediction model to obtain a trained commodity information prediction model so as to
  • the obtained product keyword information in the product search request is input into the trained product information prediction model to obtain multiple product introduction text information that matches the product keyword information, and the product introduction text information is respectively combined with the multiple product introduction text information.
  • Corresponding multiple product information is displayed, so as to achieve accurate product push for users' real-time needs.
  • the product information prediction model trained in this application can recommend product information that matches the product demand text information for the user according to the product demand text information input by the user.
  • an embodiment of the present application provides a product information push device based on a product information prediction model.
  • the device includes: a building module 31, a training module 32, and a matching module 34.
  • the display module 35 As shown in FIG. 3, the device includes: a building module 31, a training module 32, and a matching module 34.
  • the display module 35 As shown in FIG. 3, the device includes: a building module 31, a training module 32, and a matching module 34.
  • the construction module 31 may be used to construct a data candidate set according to the acquired text information of commodity requirements
  • the training module 32 can be used to train the initial product information prediction model by using the constructed data candidate set to obtain a trained product information prediction model
  • the matching module 34 may be used to input the obtained product keyword information in the product search request into a trained product information prediction model to obtain multiple product introduction text information matching the product keyword information;
  • the display module 35 may be used to respectively display a plurality of commodity information corresponding to the plurality of commodity introduction text information.
  • a monitoring module 33 is also included.
  • the commodity requirement text information includes commodity keyword information and multiple commodity introduction text information corresponding to the commodity keyword information; among them, the commodity requirement text information used to construct the data candidate set
  • the multiple product introduction text information corresponding to the product keyword information includes the first product introduction text information of the product that the user has purchased and the second product introduction text information of the product that the user has not purchased.
  • the first product introduction text information in the data candidate set is a positive sample
  • the second product introduction text information is a negative sample.
  • the training module 32 specifically includes: a first calculation unit 321, a second Calculating unit 322 and minimizing training unit 323.
  • the first calculation unit 321 may be configured to calculate the distance similarity between the commodity keyword information and the first commodity introduction text information and a plurality of second commodity introduction text information to obtain multiple similarity values.
  • the second calculation unit 322 may be configured to calculate the similarity probability between the commodity keyword information and the first commodity introduction text information according to the multiple similarity values.
  • the minimization training unit 323 may be used to train the initial product information prediction model according to the similarity probability to obtain a trained product information prediction model.
  • the minimization training unit 323 can be specifically used to determine the network parameters of the product information prediction model by performing maximum likelihood estimation on the similarity probability; according to the determined product information prediction model network Parameters to obtain a trained product information prediction model.
  • the monitoring module 33 may be used to obtain the product keyword information in the product search request when the product search request from the user is monitored.
  • the matching module 34 specifically includes: a first acquisition unit 341, a second acquisition unit 342, and a similarity calculation unit 343.
  • the first obtaining unit 341 may be used to obtain multiple commodity information matching the commodity keyword information by using the trained commodity information prediction model;
  • the second obtaining unit 342 may be configured to obtain multiple product introduction text information corresponding to the multiple product information according to the multiple product information obtained by matching;
  • the similarity calculation unit 343 may be configured to calculate the similarity between the product keyword information and the obtained multiple product introduction text information to obtain multiple product introduction text information that matches the product keyword information.
  • the similarity calculation unit 343 may be specifically configured to calculate the distance similarity between the semantic vector corresponding to the product keyword information and the semantic vector corresponding to the acquired multiple product introduction text information, to obtain multiple A similarity value; sorting the calculated similarity values in descending order to obtain a descending sorting result; according to the preset product matching value, according to the descending sorting result to obtain a plurality of products matching the product keyword information Introduce text information.
  • an embodiment of the present application also provides a non-volatile readable storage medium on which computer readable instructions are stored, and the program is executed when the processor is executed.
  • the commodity information push method based on the commodity information prediction model as shown in Figs. 1 and 2 above.
  • the technical solution of the present application can be embodied in the form of a software product, and the software product can be stored in a non-volatile non-volatile readable storage medium (can be CD-ROM, U disk, mobile hard disk) Etc.), including several instructions to enable a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in each implementation scenario of this application.
  • a computer device which may be a personal computer, a server, or a network device, etc.
  • the embodiments of the present application also provide a computer device, which can be a personal computer, a server, or a network.
  • the physical device includes a non-volatile readable storage medium and a processor; the non-volatile readable storage medium is used to store computer readable instructions; and the processor is used to execute computer readable instructions to achieve the above
  • the commodity information push method based on the commodity information prediction model shown in FIG. 1 and FIG. 2.
  • the computer device may also include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a Wi-Fi module, and so on.
  • the user interface may include a display screen (Display), an input unit such as a keyboard (Keyboard), etc., and the optional user interface may also include a USB interface, a card reader interface, and the like.
  • the network interface can optionally include a standard wired interface, a wireless interface (such as a Bluetooth interface, a WI-FI interface), etc.
  • the non-volatile readable storage medium may also include an operating system and a network communication module.
  • the operating system is a program that manages the hardware and software resources of computer equipment, and supports the operation of information processing programs and other software and/or programs.
  • the network communication module is used to implement communication between various components in the non-volatile readable storage medium and communication with other hardware and software in the physical device.
  • this application can be implemented by means of software plus a necessary general hardware platform, or by hardware.
  • this embodiment can recommend and recommend the product demand text for the user based on the product demand text information input by the user through the trained product information prediction model
  • Product information with a high degree of information matching does not have the cold start problem in the prior art, and does not require manual input of product information in the process of network model construction, which can ensure that the entire process of product information prediction model construction is automated.

Abstract

A commodity information prediction model-based commodity information pushing method and apparatus, a non-volatile readable storage medium, and a computer device, relating to the technical field of image recognition and capable of improving image recognition accuracy. The method comprises: constructing a data candidate set according to obtained commodity demand text information (101); training an initialized commodity information prediction model by using the constructed data candidate set to obtain a trained commodity information prediction model (102); inputting commodity keyword information in an obtained commodity search request into the trained commodity information prediction model to obtain multiple pieces of commodity introduction text information matching the commodity keyword information (103); and separately displaying multiple pieces of commodity information corresponding to the multiple pieces of commodity introduction text information (104). The method is applicable to accurate pushing of e-commerce commodities in the mobile internet and the O2O mode.

Description

基于商品信息预测模型的商品信息推送方法及装置、非易失性可读存储介质、计算机设备Commodity information push method and device based on commodity information prediction model, non-volatile readable storage medium, and computer equipment
本申请要求与2019年8月22日提交中国专利局、申请号为201910780785.1、申请名称为“商品信息推送方法及装置、存储介质、计算机设备”的中国专利申请的优先权，其全部内容通过引用结合在申请中。This application claims the priority of the Chinese patent application filed on August 22, 2019 with the Chinese Patent Office, the application number is 201910780785.1, and the application name is "Product information push method and device, storage medium, computer equipment", the entire content of which is incorporated by reference Incorporate in the application.
技术领域Technical field
本申请涉及信息处理技术领域，尤其是涉及到基于商品信息预测模型的商品信息推送方法及装置、非易失性可读存储介质及计算机设备。This application relates to the field of information processing technology, in particular to a commodity information push method and device based on a commodity information prediction model, a non-volatile readable storage medium, and computer equipment.
背景技术Background technique
随着移动互联网和O2O模式的迅猛发展，电商的商品类别越来越丰富，数据信息量也随之增长，数据挖掘已成为电商提高效益的主要方式。通过数据挖掘将用户和商品相结合，能够为不同的用户推送不同的商品，以满足不同的用户对于商品的不同需求，从而提升用户的购物体验。With the rapid development of the mobile Internet and O2O model, e-commerce product categories are becoming more and more abundant, and the amount of data information has also increased. Data mining has become the main way for e-commerce to improve efficiency. Combining users and commodities through data mining can push different commodities to different users to meet different users' different needs for commodities, thereby enhancing users' shopping experience.
现有根据用户信息推荐商品的方法主要包括两种：1)基于用户历史购买行为信息，为用户推荐相应的商品；2)基于商品自身的可量化属性信息，为用户推荐相应的商品。基于用户历史购买行为信息为用户推荐相应的商品，存在冷启动的问题；基于商品自身的可量化属性信息为用户推荐相应的商品，在网络模型构建的过程中需要通过手动输入商品信息以实现对用户的商品推送，不适用于海量商品的情况，且上述两种方法均无法根据用户的实时需求来实现精准的推送功能。The existing methods for recommending products based on user information mainly include two: 1) recommending corresponding products for users based on user historical purchasing behavior information; 2) recommending corresponding products for users based on quantifiable attribute information of the products themselves. Recommending corresponding products to users based on user historical purchasing behavior information has the problem of cold start; recommending corresponding products to users based on the quantifiable attribute information of the products themselves. In the process of building the network model, manual input of product information is required to achieve The user's product push is not suitable for mass products, and neither of the above two methods can achieve accurate push functions based on the user's real-time needs.
发明内容Summary of the invention
有鉴于此，本申请提供了基于商品信息预测模型的商品信息推送方法及装置、非易失性可读存储介质、计算机设备，主要目的在于解决现有商品信息的推送方法无法根据用户的实时需求精准推送商品信息的技术问题。In view of this, this application provides a product information push method and device based on a product information prediction model, non-volatile readable storage media, and computer equipment. The main purpose is to solve the problem that existing product information push methods cannot meet the real-time needs of users. The technical problem of accurately pushing product information.
根据本申请的一个方面，提供了一种基于商品信息预测模型的商品信息推送方法，该方法包括：According to one aspect of the present application, there is provided a commodity information push method based on a commodity information prediction model, the method including:
根据获取到的商品需求文本信息构建数据候选集；Construct a data candidate set according to the acquired text information of commodity requirements;
利用所构建的数据候选集对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型；Use the constructed data candidate set to train the initial product information prediction model to obtain a trained product information prediction model;
将获取到的商品搜索请求中的商品关键词信息输入训练好的商品信息预测模型，得到与所述商品关键词信息匹配的多个商品介绍文本信息；Input the acquired product keyword information in the product search request into the trained product information prediction model to obtain multiple product introduction text information matching the product keyword information;
分别将与所述多个商品介绍文本信息对应的多个商品信息进行显示。A plurality of commodity information corresponding to the plurality of commodity introduction text information are respectively displayed.
根据本申请的另一方面，提供了一种基于商品信息预测模型的商品信息推送装置，该装置包括：According to another aspect of the present application, there is provided a product information push device based on a product information prediction model, the device including:
构建模块，用于根据获取到的商品需求文本信息构建数据候选集；The construction module is used to construct a data candidate set according to the acquired text information of commodity requirements;
训练模块，用于利用所构建的数据候选集对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型；The training module is used to train the initial product information prediction model by using the constructed data candidate set to obtain the trained product information prediction model;
匹配模块，用于将获取到的商品搜索请求中的商品关键词信息输入训练好的商品信息预测模型，得到与所述商品关键词信息匹配的多个商品介绍文本信息；The matching module is used to input the acquired product keyword information in the product search request into the trained product information prediction model to obtain multiple product introduction text information that matches the product keyword information;
显示模块，用于分别将与所述多个商品介绍文本信息对应的多个商品信息进行显示。The display module is used to respectively display a plurality of commodity information corresponding to the plurality of commodity introduction text information.
依据本申请又一个方面，提供了一种非易失性可读存储介质，其上存储有计算机可读指令，所述程序被处理器执行时实现上述基于商品信息预测模型的商品信息推送方法。According to yet another aspect of the present application, there is provided a non-volatile readable storage medium having computer readable instructions stored thereon, and when the program is executed by a processor, the foregoing method for pushing commodity information based on a commodity information prediction model is realized.
依据本申请再一个方面，提供了一种计算机设备，包括非易失性可读存储介质、处理器及存储在非易失性可读存储介质上并可在处理器上运行的计算机可读指令，所述处理器执行所述程序时实现上述基于商品信息预测模型的商品信息推送方法。According to another aspect of the present application, a computer device is provided, including a non-volatile readable storage medium, a processor, and computer readable instructions stored on the non-volatile readable storage medium and running on the processor When the processor executes the program, the foregoing method for pushing commodity information based on the commodity information prediction model is implemented.
借由上述技术方案，本申请提供的基于商品信息预测模型的商品信息推送方法及装置、非易失性可读存储介质、计算机设备，与现有的商品信息推送的技术方案相比，本申请根据获取到的商品需求文本信息构建数据候选集，并利用所构建的数据候选集对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型，以便将获取到的商品搜索请求中的商品关键词信息输入训练好的商品信息预测模型，得到与所述商品关键词信息匹配的多个商品介绍文本信息，并分别将与所述多个商品介绍文本信息对应的多个商品信息进行显示，从而对用户的实时需求实现精准的商品推送。可见，通过训练好的商品信息预测模型能够根据用户输入的商品需求文本信息，为用户推荐与商品需求文本信息匹配度较高的商品信息，不存在现有技术中的冷启动问题，且不需要在网络模型构建的过程中人工输入商品信息，能够保证商品信息预测模型构建的整个过程自动化完成。With the above technical solutions, the commodity information push method and device based on the commodity information prediction model, non-volatile readable storage medium, and computer equipment provided by this application are compared with the existing technical solutions for commodity information push. Construct a data candidate set according to the acquired text information of commodity demand, and use the constructed data candidate set to train the initial commodity information prediction model to obtain the trained commodity information prediction model so as to search for the commodities in the obtained commodity search request. The keyword information is input to the trained product information prediction model to obtain multiple product introduction text information matching the product keyword information, and to display multiple product information corresponding to the multiple product introduction text information respectively. In this way, accurate product pushes can be achieved for users' real-time needs. It can be seen that the trained product information prediction model can recommend the product information with a high degree of matching with the product demand text information according to the product demand text information input by the user. There is no cold start problem in the prior art, and there is no need Manual input of product information in the process of building the network model can ensure that the entire process of building the product information prediction model is completed automatically.
上述说明仅是本申请技术方案的概述，为了能够更清楚了解本申请的技术手段，而可依照说明书的内容予以实施，并且为了让本申请的上述和其它目的、特征和优点能够更明显易懂，以下特举本申请的具体实施方式。The above description is only an overview of the technical solution of this application. In order to understand the technical means of this application more clearly, it can be implemented in accordance with the content of the specification, and to make the above and other purposes, features and advantages of this application more obvious and understandable. , The following specifically cite the specific implementation of this application.
附图说明Description of the drawings
此处所说明的附图用来提供对本申请的进一步理解，构成本申请的一部分，本申请的示意性实施例及其说明用于解释本申请，并不构成对本申请的不当限定。在附图中：The drawings described here are used to provide a further understanding of the application and constitute a part of the application. The exemplary embodiments and descriptions of the application are used to explain the application and do not constitute an improper limitation of the application. In the attached picture:
图1示出了本申请实施例提供的一种基于商品信息预测模型的商品信息推送方法的流程示意图；FIG. 1 shows a schematic flowchart of a method for pushing commodity information based on a commodity information prediction model provided by an embodiment of the present application;
图2示出了本申请实施例提供的另一种基于商品信息预测模型的商品信息推送方法的流程示意图；FIG. 2 shows a schematic flowchart of another method for pushing commodity information based on a commodity information prediction model provided by an embodiment of the present application;
图3示出了本申请实施例提供的一种基于商品信息预测模型的商品信息推送装置的结构示意图。Fig. 3 shows a schematic structural diagram of a product information push device based on a product information prediction model provided by an embodiment of the present application.
具体实施方式detailed description
下文中将参考附图并结合实施例来详细说明本申请。需要说明的是，在不冲突的情况下，本申请中的实施例及实施例中的特征可以相互组合。Hereinafter, the application will be described in detail with reference to the drawings and in conjunction with embodiments. It should be noted that the embodiments in this application and the features in the embodiments can be combined with each other if there is no conflict.
针对现有基于用户历史购买行为信息为用户推荐相应的商品，存在冷启动的技术问题，基于商品自身的可量化属性信息为用户推荐相应的商品，在网络模型构建的过程中需要通过手动输入商品信息以实现对用户的商品推送，不适用于海量商品的技术问题，以及无法根据用户的实时需求来实现精准的推送功能的技术问题。本实施例提供了一种基于商品信息预测模型的商品信息推送方法，能够有效避免现有技术中冷 启动的问题，以及网络模型构建不够自动化、智能化的问题，同时进一步提升了商品信息推荐的准确度，如图1所示，该方法包括：In view of the existing technical problems of cold start for users to recommend corresponding products based on user historical purchasing behavior information, users are recommended corresponding products based on the quantifiable attribute information of the products themselves. Manual input of products is required in the process of network model construction Information is used to realize product push to users, which is not suitable for the technical problems of massive goods, and technical problems that cannot realize accurate push functions according to the real-time needs of users. This embodiment provides a product information push method based on a product information prediction model, which can effectively avoid the problem of cold start in the prior art and the problem of insufficient automation and intelligence in the construction of network models, and at the same time further improve the recommendation of product information. Accuracy, as shown in Figure 1, the method includes:
101、根据获取到的商品需求文本信息构建数据候选集。101. Construct a data candidate set according to the acquired text information of commodity requirements.
在本实施例中，收集用户的历史购买记录信息，并获取针对不同商品的商品需求文本信息作为数据样本，对获取到的属于同一商品的多个商品需求文本信息进行标记，从而得到用于训练商品信息预测模型的数据候选集。In this embodiment, the user’s historical purchase record information is collected, and the product demand text information for different products is obtained as a data sample, and the acquired multiple product demand text information belonging to the same product is marked, so as to obtain the text information for training The data candidate set of the product information prediction model.
102、利用所构建的数据候选集对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型。102. Use the constructed data candidate set to train the initial commodity information prediction model to obtain a trained commodity information prediction model.
在本实施例中，基于神经网络模型(例如，卷积神经网络模型、循环神经网络模型或者概率神经网络模型)构建初始化商品推送预测模型，初始化商品推送预测模型的框架具体为：In this embodiment, the initial product push prediction model is constructed based on a neural network model (for example, a convolutional neural network model, a recurrent neural network model, or a probabilistic neural network model), and the framework of the initial product push prediction model is specifically:
第一层：输入层，对输入信息进行格式处理，输入信息为数据候选集；The first layer: the input layer, format the input information, and the input information is the data candidate set;
第二层：编码层，对格式处理后的输入信息以字符为单位进行one-hot编码，得到相应的语义向量；The second layer: encoding layer, which performs one-hot encoding on the input information after format processing in units of characters to obtain the corresponding semantic vector;
第三层至第六层：映射层，语义向量经由多个包含激活函数tanh的全连接层处理后，得到相应的多个输出结果；The third to sixth layers: the mapping layer, the semantic vector is processed by multiple fully connected layers containing the activation function tanh to obtain corresponding multiple output results;
第七层：匹配层，将得到的多个输出结果进行相似度计算，得到多个相似度计算结果；The seventh layer: the matching layer, which performs similarity calculations on multiple output results to obtain multiple similarity calculation results;
第八层：Softmax层，利用激活函数Softmax对计算得到的多个相似度计算结果进行归一化处理，得到相似度概率。The eighth layer: the Softmax layer, the activation function Softmax is used to normalize the calculated similarity calculation results to obtain the similarity probability.
基于所构建的数据候选集对上述初始化商品推送预测模型进行训练，得到训练好的商品信息预测模型，以便利用训练好的商品信息预测模型，根据来自用户的商品搜索请求实现商品信息预测，从而得到满足用户商品搜索需求的商品信息。Based on the constructed data candidate set, train the above-mentioned initial product push prediction model to obtain a trained product information prediction model, so that the trained product information prediction model can be used to realize product information prediction according to the product search request from the user, thereby obtaining Product information that meets the user's product search needs.
103、将获取到的商品搜索请求中的商品关键词信息输入训练好的商品信息预测模型，得到与所述商品关键词信息匹配的多个商品介绍文本信息。103. Input the acquired product keyword information in the product search request into a trained product information prediction model to obtain multiple product introduction text information matching the product keyword information.
在本实施例中，根据获取到的商品搜索请求中的商品关键词信息得到与该商品关键词信息相匹配的多个商品介绍文本信息，利用训练好的商品信息预测模型中的第一层至第六层得到用于相似度计算的商品介绍文本信息，根据第六层输出的映射结果经由第七层、第八层进行计算，得到与商品关键词信息匹配的多个商品介绍文本信息。In this embodiment, according to the obtained product keyword information in the product search request, multiple product introduction text information that matches the product keyword information are obtained, and the first layer to the first layer of the trained product information prediction model are used. The sixth layer obtains the product introduction text information used for similarity calculation, and calculates through the seventh and eighth layers according to the mapping result output by the sixth layer to obtain multiple product introduction text information matching the product keyword information.
104、分别将与所述多个商品介绍文本信息对应的多个商品信息进行显示。104. Respectively display multiple product information corresponding to the multiple product introduction text information.
在本实施例中，获取多个商品介绍文本信息对应的多个商品信息，将获取到的多个商品信息按照相似度进行降序排列并进行展示，以便用户进行浏览。In this embodiment, multiple product information corresponding to multiple product introduction text information is acquired, and the multiple acquired product information is sorted in descending order of similarity and displayed for users to browse.
对于本实施例可以按照上述方案，根据获取到的商品需求文本信息构建数据候选集，并利用所构建的数据候选集对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型，以便将获取到的商品搜索请求中的商品关键词信息输入训练好的商品信息预测模型，得到与所述商品关键词信息匹配的多个商品介绍文本信息，并分别将与所述多个商品介绍文本信息对应的多个商品信息进行显示，从而对用户的实 时需求实现精准的商品推送。与现有的商品信息推送的技术方案相比，本申请通过训练好的商品信息预测模型能够根据用户输入的商品需求文本信息，为用户推荐与商品需求文本信息匹配度较高的商品信息，不存在现有技术中的冷启动问题，且不需要在网络模型构建的过程中人工输入商品信息，能够保证商品信息预测模型构建的整个过程自动化完成。For this embodiment, according to the above scheme, a data candidate set can be constructed according to the obtained product demand text information, and the constructed data candidate set can be used to train the initial product information prediction model to obtain a trained product information prediction model, so as to The obtained product keyword information in the product search request is input into the trained product information prediction model to obtain multiple product introduction text information that matches the product keyword information, and the product introduction text information is respectively combined with the multiple product introduction text information. Corresponding multiple product information is displayed, so as to achieve accurate product push for users' real-time needs. Compared with the existing technical solutions for pushing product information, the product information prediction model trained in this application can recommend product information that matches the product demand text information for the user according to the product demand text information input by the user. There is a cold start problem in the prior art, and there is no need to manually input product information in the process of building a network model, which can ensure that the entire process of building a product information prediction model is completed automatically.
进一步的，作为上述实施例具体实施方式的细化和扩展，为了完整说明本实施例的具体实施过程，提供了另一种基于商品信息预测模型的商品信息推送方法，如图2所示，该方法包括：Further, as a refinement and extension of the specific implementation of the foregoing embodiment, in order to fully explain the specific implementation process of this embodiment, another method for pushing product information based on a product information prediction model is provided, as shown in FIG. Methods include:
201、根据获取到的商品需求文本信息构建数据候选集。201. Construct a data candidate set according to the acquired text information of commodity requirements.
为了说明步骤201的具体实施方式，作为一种优选实施例，所述商品需求文本信息包括商品关键词信息，以及对应所述商品关键词信息的多个商品介绍文本信息；其中，用于构建数据候选集的商品需求文本信息中的对应商品关键词信息的多个商品介绍文本信息包括用户购买过的商品的第一商品介绍文本信息和用户未购买过的商品的第二商品介绍文本信息。To illustrate the specific implementation of step 201, as a preferred embodiment, the commodity demand text information includes commodity keyword information, and multiple commodity introduction text information corresponding to the commodity keyword information; wherein, it is used to construct data The multiple product introduction text information corresponding to the product keyword information in the product demand text information of the candidate set includes the first product introduction text information of the product that the user has purchased and the second product introduction text information of the product that the user has not purchased.
在本实施例中，商品需求文本信息包括：商品关键词信息(例如，输入的商品名称、商品名称关键词)，以及每个商品名称关键词对应的多个具有相同商品名称或者商品名称关键词的商品介绍文本信息。对具有同样商品名称或者商品名称关键词的商品介绍文本信息进行标记，即将用户购买过的商品的第一商品介绍文本信息标记为正样本，将用户未购买过的商品的第二商品介绍文本标记为负样本，并作为数据候选集。In this embodiment, the commodity demand text information includes: commodity keyword information (for example, the input commodity name, commodity name keywords), and each commodity name keyword corresponds to multiple keywords with the same commodity name or commodity name The product introduction text information. Mark the product introduction text information with the same product name or product name keywords, that is, mark the first product introduction text information of the product that the user has purchased as a positive sample, and mark the second product introduction text of the product that the user has not purchased It is a negative sample and serves as a data candidate set.
202、所述数据候选集中的第一商品介绍文本信息为正样本，所述第二商品介绍文本信息为负样本，将所述商品关键词信息分别与第一商品介绍文本信息和多个第二商品介绍文本信息进行距离相似度计算，得到多个相似度值。202. The first product introduction text information in the data candidate set is a positive sample, and the second product introduction text information is a negative sample, and the product keyword information is separated from the first product introduction text information and multiple second The product introduction text information is calculated by distance similarity to obtain multiple similarity values.
在本实施例中，利用初始化商品信息预测模型的第一层(输入层)对输入信息进行预处理，输入信息为数据候选集，数据候选集中商品需求文本信息中的商品名称关键词的字符数量不得超过30个字符，具有相同商品名称或者商品名称关键词的商品介绍文本信息的字符数量不得超过500个字符。若字符数量少于所要求的字符数量，则在商品名称关键词，以及具有相同商品名称或者商品名称关键词的商品介绍文本的末端补0；若字符数量多于所要求的字符数量，则将商品名称关键词，以及具有相同商品名称或者商品名称关键词的商品介绍文本的超出部分直接舍弃。In this embodiment, the first layer (input layer) of the initial product information prediction model is used to preprocess the input information. The input information is a data candidate set. The number of characters of the product name keywords in the product demand text information in the data candidate set No more than 30 characters, and the number of characters in the product introduction text information with the same product name or product name keywords should not exceed 500 characters. If the number of characters is less than the required number of characters, add 0 to the end of the product name keywords and the product introduction text with the same product name or product name keywords; if the number of characters is more than the required number of characters, the Product name keywords, and the excess part of the product introduction text with the same product name or product name keywords are directly discarded.
其中，对商品需求文本信息中商品介绍文本信息的样本标记具体为，用户购买过的商品对应的第一商品介绍文本信息为正样本，默认对应一个商品关键词信息，以及用户购买过的商品信息；用户未购买过的多个商品对应的多个第二商品介绍文本信息为多个负样本，默认对应一个商品关键词信息，以及多个商品信息。Among them, the sample mark of the product introduction text information in the product demand text information is specifically that the first product introduction text information corresponding to the product purchased by the user is a positive sample, which corresponds to a product keyword information by default, and the product information purchased by the user ; The multiple second product introduction text information corresponding to multiple products that the user has not purchased are multiple negative samples, which correspond to one product keyword information and multiple product information by default.
相应地，利用初始化商品信息预测模型的第二层(编码层)对预处理后的商品需求文本信息以字符为单位进行one-hot编码，即对商品名称或者名称关键词，属于正样本的第一商品介绍文本，以及多个属于负样本的第二商品介绍文本进行编码，分别得到对应的语义向量。其中，one-hot编码是指使用N位状态寄存器来对N个状态进行编码，每个状态都有独立的寄存器位，且在任意时候保证其中只有一个状态有效。 例如，对“青岛特产”进行分词得到特征词“青岛”、“特产”，即“青岛”索引为1，“特产”索引为2，得到每个特征词的语义向量为：青岛(10)，特产(01)。Correspondingly, the second layer (coding layer) of the initial product information prediction model is used to perform one-hot encoding on the preprocessed product demand text information in character units, that is, the product name or name keyword belongs to the first sample of the positive sample. One product introduction text and multiple second product introduction texts belonging to negative samples are encoded to obtain corresponding semantic vectors respectively. Among them, one-hot encoding refers to the use of N-bit status registers to encode N states, each state has an independent register bit, and it is guaranteed that only one state is valid at any time. For example, by segmenting "Qingdao special product", the feature words "Qingdao" and "special product" are obtained, that is, the index of "Qingdao" is 1, and the index of "special product" is 2, and the semantic vector of each feature word is: Qingdao (10), Specialties (01).
在实际应用场景中，将每组样本中的商品名称或者名称关键词、第一商品介绍文本、多个第二商品介绍文本进行one-hot编码，得到对应的语义向量，即商品名称或者名称关键词的语义向量记为Q，属于正样本的第一商品介绍文本的语义向量记为D 1，以及多个属于负样本的第二商品介绍文本的语义向量记为D 2-D n。 In actual application scenarios, one-hot encode the product name or name keyword, the first product introduction text, and multiple second product introduction texts in each set of samples to obtain the corresponding semantic vector, that is, the product name or name key The semantic vector of the word is denoted as Q, the semantic vector of the first product introduction text belonging to the positive sample is denoted as D 1 , and the semantic vector of multiple second product introduction texts belonging to the negative sample is denoted as D 2 -D n .
相应地，利用初始化商品信息预测模型的第三层至第六层(映射层)，将得到的语义向量经由多个包含激活函数tanh的全连接层，得到用于表征商品关键词信息的输出结果，以及用于表征商品介绍文本信息的输出结果。Correspondingly, using the third layer to the sixth layer (mapping layer) of the initial product information prediction model, the obtained semantic vector is passed through multiple fully connected layers containing the activation function tanh to obtain the output result for characterizing the product keyword information , And the output result used to characterize the text information of the product introduction.
其中，第三层的输入数据为第二层的输出数据(Q，D 1，D 2，……，D n)，随机初始化参数矩阵为W，b，根据输入数据、随机初始化参数矩阵和激活函数进行计算，得到输出结果，具体的计算公式为： Among them, the input data of the third layer is the output data of the second layer (Q, D 1 , D 2 ,..., D n ), and the random initialization parameter matrix is W, b. According to the input data, the random initialization parameter matrix and activation The function calculates and obtains the output result. The specific calculation formula is:
Figure PCTCN2019118488-appb-000001
Figure PCTCN2019118488-appb-000001
第四层的随机初始化参数矩阵为W 1，b 1，具体的计算公式为： The random initialization parameter matrix of the fourth layer is W 1 , b 1 , and the specific calculation formula is:
Figure PCTCN2019118488-appb-000002
Figure PCTCN2019118488-appb-000002
以此推类，第五层、第六层与第三层、第四层相同。By analogy, the fifth and sixth floors are the same as the third and fourth floors.
在本实施例中，利用初始化商品信息预测模型的第七层(匹配层)，将得到的用于表征商品关键词信息的输出结果，分别与用于表征商品介绍文本信息的输出结果进行距离相似度计算，得到多个相似度值。具体的计算公式为：In this embodiment, using the seventh layer (matching layer) of the initial product information prediction model, the obtained output results used to characterize product keyword information are respectively similar to the output results used to characterize product introduction text information. Degree calculation to obtain multiple similarity values. The specific calculation formula is:
Figure PCTCN2019118488-appb-000003
Figure PCTCN2019118488-appb-000003
其中，R(Q，D)表示用于表征商品关键词信息的输出结果 yQ与用于表征商品介绍文本信息的输出结果yD之间的相似度值，cosine(y Q，y D)表示两个语义向量的cosine距离，即语义相似性。 Among them, R(Q, D) represents the similarity value between the output result yQ used to characterize the product keyword information and the output result yD used to characterize the product introduction text information, and cosine(y Q , y D ) represents two The cosine distance of the semantic vector is the semantic similarity.
203、根据所述多个相似度值，计算得到所述商品关键词信息与第一商品介绍文本信息的相似度概率。203. According to the multiple similarity values, calculate the similarity probability between the commodity keyword information and the first commodity introduction text information.
在本实施例中，利用初始化商品信息预测模型的第八层(Softmax层)，中的激活函数Softmax对计算得到的多个相似度值进行归一化处理，得到商品关键词信息Q与第一商品介绍文本信息D 1的相似度概率。具体的计算公式为： In this embodiment, the activation function Softmax in the eighth layer (Softmax layer) of the initial product information prediction model is used to normalize the calculated similarity values to obtain the product keyword information Q and the first The similarity probability of the product introduction text information D 1. The specific calculation formula is:
Figure PCTCN2019118488-appb-000004
Figure PCTCN2019118488-appb-000004
其中，P(D 1|Q)表示商品关键词信息Q与第一商品介绍文本信息D1的相似度概率，γ为权重参数，即Softmax的平滑因子。 Among them, P(D 1 |Q) represents the probability of similarity between the product keyword information Q and the first product introduction text information D1, and γ is the weight parameter, that is, the smoothing factor of Softmax.
204、根据所述相似度概率对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型。204. Train the initial commodity information prediction model according to the similarity probability to obtain a trained commodity information prediction model.
为了说明步骤204的具体实施方式，作为一种优选实施例，步骤204具体可以包括：通过对所述相似度概率进行极大似然估计，确定商品信息预测模型的网络参数；根据所确定的商品信息预测模型的网络参数，得到训练好的商品信息预测模型。To illustrate the specific implementation of step 204, as a preferred embodiment, step 204 may specifically include: determining the network parameters of the product information prediction model by performing maximum likelihood estimation on the similarity probability; and according to the determined product The network parameters of the information prediction model are used to obtain a trained commodity information prediction model.
在本实施例中，根据商品关键词信息Q与第一商品介绍文本信息D 1的相似度概率确定损失函数，该损失函数为对数似然损失函数，即基于所有商品关键词信息Q与第一商品介绍文本信息D 1的相似度概率的极大似然估计对初始化商品推送预测模型进行训练，得到训练好的商品推送预测模型，以使商品关键词信息Q匹配第一商品介绍文本信息D 1的概率最大化，损失函数最小化，损失函数具体的计算公式为： In the present embodiment, the loss function is determined based on the similarity probability Q first keyword information commodity product description text message D 1, and the loss function for the log-likelihood function loss, i.e. all goods based on the first key data Q 1. The maximum likelihood estimation of the similarity probability of the product introduction text information D 1 Train the initial product push prediction model to obtain the trained product push prediction model so that the product keyword information Q matches the first product introduction text information D The probability of 1 is maximized, and the loss function is minimized. The specific calculation formula of the loss function is:
Figure PCTCN2019118488-appb-000005
Figure PCTCN2019118488-appb-000005
205、当监测到来自用户的商品搜索请求时，获取所述商品搜索请求中的商品关键词信息。205. When a product search request from a user is monitored, obtain product keyword information in the product search request.
206、利用训练好的商品信息预测模型，获取与所述商品关键词信息相匹配的多个商品信息。206. Use the trained product information prediction model to obtain multiple product information that matches the product keyword information.
207、根据匹配得到的多个商品信息，获取与所述多个商品信息对应的多个商品介绍文本信息。207. Acquire multiple product introduction text information corresponding to the multiple product information according to the multiple product information obtained by the matching.
在本实施例中，当接收到来自用户的商品搜索请求时，获取商品搜索请求中的商品关键词信息，以及与商品关键词匹配的多个商品介绍文本信息。将获取到的商品关键词信息和多个商品介绍文本信息分别输入训练好的商品推送预测模型的商品关键词模块(即，商品关键词Q对应的子网络模型)，以及正样本模块(即，第一商品介绍文本信息D 1对应的子网络模型)，得到商品关键词的语义向量Q，以及多个商品介绍文本信息的语义向量D。 In this embodiment, when a product search request from a user is received, the product keyword information in the product search request and multiple product introduction text information matching the product keywords are acquired. Input the acquired product keyword information and multiple product introduction text information into the product keyword module of the trained product push prediction model (ie, the sub-network model corresponding to the product keyword Q) and the positive sample module (ie, the first product description text message D 1 corresponding to a sub-network model), to give the product keywords semantic vector Q, and a plurality of semantic vector product description text information D.
在实际应用场景中，对获取到的商品关键词和多个商品介绍文本信息进行字符处理，得到统一字符数量的商品关键词和多个商品介绍文本信息，对统一后的商品关键词利用商品推送预测模型中的商品关键词模块进行编码，以及对统一后的多个商品介绍文本信息利用商品推送预测模型中的正样本模块进行编码，得到用于表征商品关键词的语义向量Q，以及多个商品介绍文本信息的语义向量D。In the actual application scenario, character processing is performed on the obtained product keywords and multiple product introduction text information to obtain a product keyword with a uniform number of characters and multiple product introduction text information, and use product push for the unified product keywords The product keyword module in the prediction model is encoded, and the unified product introduction text information is encoded using the positive sample module in the product push prediction model to obtain the semantic vector Q used to represent the product keywords, and multiple The semantic vector D of the product introduction text information.
208、将所述商品关键词信息分别与获取到的多个商品介绍文本信息进行相似度计算，得到与所述商品关键词信息匹配的多个商品介绍文本信息。208. Perform similarity calculations on the product keyword information and the obtained multiple product introduction text information respectively, to obtain multiple product introduction text information matching the product keyword information.
为了说明步骤208的具体实施方式，作为一种优选实施例，步骤208具体可以包括：将所述商品关键词信息对应的语义向量分别与获取到的多个商品介绍文本信息对应的语义向量进行距离相似度计算，得到多个相似度值；对计算得到的多个相似度值进行降序排列，得到降序排列结果；按照预设的商品匹配值，根据所述降序排列结果得到与所述商品关键词信息匹配的多个商品介绍文本信息。其中，预设的商品匹配值可以表示为超过相似度90％的所有商品，也可以表示为降序排列结果中预设排名(例如，TOP100)内的商品，此处不对预设的商品匹配值的设定标准和设定维度进行具体限定。To illustrate the specific implementation of step 208, as a preferred embodiment, step 208 may specifically include: distance the semantic vectors corresponding to the product keyword information from the semantic vectors corresponding to the acquired multiple product introduction text information. Similarity calculation to obtain multiple similarity values; to sort the calculated similarity values in descending order to obtain a descending sorting result; according to the preset merchandise matching value, according to the descending sorting result to obtain the keywords with the merchandise Multiple product introduction text information with matching information. Among them, the preset product matching value can be expressed as all the products with more than 90% similarity, or it can be expressed as the products in the preset ranking (for example, TOP100) in the descending order result. There is no matching value for the preset product here. Set standards and set dimensions for specific restrictions.
209、分别将与所述多个商品介绍文本信息对应的多个商品信息进行显示。209. Display a plurality of commodity information corresponding to the plurality of commodity introduction text information respectively.
在本实施例中，根据训练好的商品信息预测模型输出的多个商品介绍文本信息，获取分别对应该多个商品介绍文本信息的多个商品信息，并生成商品列表，即确定推荐给用户的多个商品。In this embodiment, according to the multiple product introduction text information output by the trained product information prediction model, multiple product information corresponding to the multiple product introduction text information are obtained, and a product list is generated, that is, the product list that is recommended to the user is determined Multiple products.
通过应用本实施例的技术方案，根据获取到的商品需求文本信息构建数据候选集，并利用所构建的数据候选集对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型，以便将获取到的商品搜索请求中的商品关键词信息输入训练好的商品信息预测模型，得到与所述商品关键词信息匹配的多个商品介绍文本信息，并分别将与所述多个商品介绍文本信息对应的多个商品信息进行显示，从而对用户的实时 需求实现精准的商品推送。与现有的商品信息推送的技术方案相比，本申请通过训练好的商品信息预测模型能够根据用户输入的商品需求文本信息，为用户推荐与商品需求文本信息匹配度较高的商品信息，不存在现有技术中的冷启动问题，且不需要在网络模型构建的过程中人工输入商品信息，能够保证商品信息预测模型构建的整个过程自动化完成。By applying the technical solution of this embodiment, a data candidate set is constructed based on the acquired commodity demand text information, and the constructed data candidate set is used to train the initial commodity information prediction model to obtain a trained commodity information prediction model so as to The obtained product keyword information in the product search request is input into the trained product information prediction model to obtain multiple product introduction text information that matches the product keyword information, and the product introduction text information is respectively combined with the multiple product introduction text information. Corresponding multiple product information is displayed, so as to achieve accurate product push for users' real-time needs. Compared with the existing technical solutions for pushing product information, the product information prediction model trained in this application can recommend product information that matches the product demand text information for the user according to the product demand text information input by the user. There is a cold start problem in the prior art, and there is no need to manually input product information in the process of building a network model, which can ensure that the entire process of building a product information prediction model is completed automatically.
进一步的，作为图1方法的具体实现，本申请实施例提供了一种基于商品信息预测模型的商品信息推送装置，如图3所示，该装置包括：构建模块31、训练模块32、匹配模块34、显示模块35。Further, as a specific implementation of the method in FIG. 1, an embodiment of the present application provides a product information push device based on a product information prediction model. As shown in FIG. 3, the device includes: a building module 31, a training module 32, and a matching module 34. The display module 35.
构建模块31，可以用于根据获取到的商品需求文本信息构建数据候选集；The construction module 31 may be used to construct a data candidate set according to the acquired text information of commodity requirements;
训练模块32，可以用于利用所构建的数据候选集对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型；The training module 32 can be used to train the initial product information prediction model by using the constructed data candidate set to obtain a trained product information prediction model;
匹配模块34，可以用于将获取到的商品搜索请求中的商品关键词信息输入训练好的商品信息预测模型，得到与所述商品关键词信息匹配的多个商品介绍文本信息；The matching module 34 may be used to input the obtained product keyword information in the product search request into a trained product information prediction model to obtain multiple product introduction text information matching the product keyword information;
显示模块35，可以用于分别将与所述多个商品介绍文本信息对应的多个商品信息进行显示。The display module 35 may be used to respectively display a plurality of commodity information corresponding to the plurality of commodity introduction text information.
在具体的应用场景中，还包括监测模块33。In a specific application scenario, a monitoring module 33 is also included.
在具体的应用场景中，所述商品需求文本信息包括商品关键词信息，以及对应所述商品关键词信息的多个商品介绍文本信息；其中，用于构建数据候选集的商品需求文本信息中的对应商品关键词信息的多个商品介绍文本信息包括用户购买过的商品的第一商品介绍文本信息和用户未购买过的商品的第二商品介绍文本信息。In a specific application scenario, the commodity requirement text information includes commodity keyword information and multiple commodity introduction text information corresponding to the commodity keyword information; among them, the commodity requirement text information used to construct the data candidate set The multiple product introduction text information corresponding to the product keyword information includes the first product introduction text information of the product that the user has purchased and the second product introduction text information of the product that the user has not purchased.
在具体的应用场景中，所述数据候选集中的第一商品介绍文本信息为正样本，所述第二商品介绍文本信息为负样本，训练模块32，具体包括：第一计算单元321、第二计算单元322、最小化训练单元323。In a specific application scenario, the first product introduction text information in the data candidate set is a positive sample, and the second product introduction text information is a negative sample. The training module 32 specifically includes: a first calculation unit 321, a second Calculating unit 322 and minimizing training unit 323.
第一计算单元321，可以用于将所述商品关键词信息分别与第一商品介绍文本信息和多个第二商品介绍文本信息进行距离相似度计算，得到多个相似度值。The first calculation unit 321 may be configured to calculate the distance similarity between the commodity keyword information and the first commodity introduction text information and a plurality of second commodity introduction text information to obtain multiple similarity values.
第二计算单元322，可以用于根据所述多个相似度值，计算得到所述商品关键词信息与第一商品介绍文本信息的相似度概率。The second calculation unit 322 may be configured to calculate the similarity probability between the commodity keyword information and the first commodity introduction text information according to the multiple similarity values.
最小化训练单元323，可以用于根据所述相似度概率对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型。The minimization training unit 323 may be used to train the initial product information prediction model according to the similarity probability to obtain a trained product information prediction model.
在具体的应用场景中，最小化训练单元323，可以具体用于通过对所述相似度概率进行极大似然估计，确定商品信息预测模型的网络参数；根据所确定的商品信息预测模型的网络参数，得到训练好的商品信息预测模型。In a specific application scenario, the minimization training unit 323 can be specifically used to determine the network parameters of the product information prediction model by performing maximum likelihood estimation on the similarity probability; according to the determined product information prediction model network Parameters to obtain a trained product information prediction model.
在具体的应用场景中，监测模块33，可以用于当监测到来自用户的商品搜索请求时，获取所述商品搜索请求中的商品关键词信息。In a specific application scenario, the monitoring module 33 may be used to obtain the product keyword information in the product search request when the product search request from the user is monitored.
在具体的应用场景中，匹配模块34，具体包括：第一获取单元341、第二获取单元342、相似计算单元343。In a specific application scenario, the matching module 34 specifically includes: a first acquisition unit 341, a second acquisition unit 342, and a similarity calculation unit 343.
第一获取单元341，可以用于利用训练好的商品信息预测模型，获取与所述商品关键词信息相匹配的多个商品信息；The first obtaining unit 341 may be used to obtain multiple commodity information matching the commodity keyword information by using the trained commodity information prediction model;
第二获取单元342，可以用于根据匹配得到的多个商品信息，获取与所述多个商品信息对应的多个商品介绍文本信息；The second obtaining unit 342 may be configured to obtain multiple product introduction text information corresponding to the multiple product information according to the multiple product information obtained by matching;
相似计算单元343，可以用于将所述商品关键词信息分别与获取到的多个商品介绍文本信息进行相似度计算，得到与所述商品关键词信息匹配的多个商品介绍文本信息。The similarity calculation unit 343 may be configured to calculate the similarity between the product keyword information and the obtained multiple product introduction text information to obtain multiple product introduction text information that matches the product keyword information.
在具体的应用场景中，相似计算单元343，可以具体用于将所述商品关键词信息对应的语义向量分别与获取到的多个商品介绍文本信息对应的语义向量进行距离相似度计算，得到多个相似度值；对计算得到的多个相似度值进行降序排列，得到降序排列结果；按照预设的商品匹配值，根据所述降序排列结果得到与所述商品关键词信息匹配的多个商品介绍文本信息。In a specific application scenario, the similarity calculation unit 343 may be specifically configured to calculate the distance similarity between the semantic vector corresponding to the product keyword information and the semantic vector corresponding to the acquired multiple product introduction text information, to obtain multiple A similarity value; sorting the calculated similarity values in descending order to obtain a descending sorting result; according to the preset product matching value, according to the descending sorting result to obtain a plurality of products matching the product keyword information Introduce text information.
需要说明的是，本申请实施例提供的一种基于商品信息预测模型的商品信息推送装置所涉及各功能单元的其他相应描述，可以参考图1和图2中的对应描述，在此不再赘述。It should be noted that, for other corresponding descriptions of the functional units involved in the product information push device based on the product information prediction model provided by the embodiment of the present application, please refer to the corresponding descriptions in FIG. 1 and FIG. 2, which will not be repeated here. .
基于上述如图1和图2所示方法，相应的，本申请实施例还提供了一种非易失性可读存储介质，其上存储有计算机可读指令，该程序被处理器执行时实现上述如图1和图2所示的基于商品信息预测模型的商品信息推送方法。基于这样的理解，本申请的技术方案可以以软件产品的形式体现出来，该软件产品可以存储在一个非易失性非易失性可读存储介质(可以是CD-ROM，U盘，移动硬盘等)中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施场景所述的方法。Based on the above-mentioned method shown in Figure 1 and Figure 2, correspondingly, an embodiment of the present application also provides a non-volatile readable storage medium on which computer readable instructions are stored, and the program is executed when the processor is executed. The commodity information push method based on the commodity information prediction model as shown in Figs. 1 and 2 above. Based on this understanding, the technical solution of the present application can be embodied in the form of a software product, and the software product can be stored in a non-volatile non-volatile readable storage medium (can be CD-ROM, U disk, mobile hard disk) Etc.), including several instructions to enable a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in each implementation scenario of this application.
基于上述如图1、图2所示的方法，以及图3所示的虚拟装置实施例，为了实现上述目的，本申请实施例还提供了一种计算机设备，具体可以为个人计算机、服务器、网络设备等，该实体设备包括非易失性可读存储介质和处理器；非易失性可读存储介质，用于存储计算机可读指令；处理器，用于执行计算机可读指令以实现上述如图1和图2所示的基于商品信息预测模型的商品信息推送方法。Based on the methods shown in Figures 1 and 2 and the virtual device embodiment shown in Figure 3, in order to achieve the above objectives, the embodiments of the present application also provide a computer device, which can be a personal computer, a server, or a network. The physical device includes a non-volatile readable storage medium and a processor; the non-volatile readable storage medium is used to store computer readable instructions; and the processor is used to execute computer readable instructions to achieve the above The commodity information push method based on the commodity information prediction model shown in FIG. 1 and FIG. 2.
可选的，该计算机设备还可以包括用户接口、网络接口、摄像头、射频(Radio Frequency，RF)电路，传感器、音频电路、WI-FI模块等等。用户接口可以包括显示屏(Display)、输入单元比如键盘(Keyboard)等，可选用户接口还可以包括USB接口、读卡器接口等。网络接口可选的可以包括标准的有线接口、无线接口(如蓝牙接口、WI-FI接口)等。本领域技术人员可以理解，本实施例提供的一种计算机设备结构并不构成对该实体设备的限定，可以包括更多或更少的部件，或者组合某些部件，或者不同的部件布置。Optionally, the computer device may also include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a Wi-Fi module, and so on. The user interface may include a display screen (Display), an input unit such as a keyboard (Keyboard), etc., and the optional user interface may also include a USB interface, a card reader interface, and the like. The network interface can optionally include a standard wired interface, a wireless interface (such as a Bluetooth interface, a WI-FI interface), etc. Those skilled in the art can understand that the structure of a computer device provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or arrange different components.
非易失性可读存储介质中还可以包括操作系统、网络通信模块。操作系统是管理计算机设备硬件和软件资源的程序，支持信息处理程序以及其它软件和/或程序的运行。网络通信模块用于实现非易失性可读存储介质内部各组件之间的通信，以及与该实体设备中其它硬件和软件之间通信。The non-volatile readable storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of computer equipment, and supports the operation of information processing programs and other software and/or programs. The network communication module is used to implement communication between various components in the non-volatile readable storage medium and communication with other hardware and software in the physical device.
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到本申请可以借助软件加必要的通用硬件平台的方式来实现，也可以通过硬件实现。通过应用本申请的技术方案，与现有的商品信息推送的技术方案相比，本实施例能够通过训练好的商品信息预测模型能够根据用户输入的商品需求文本信息，为用 户推荐与商品需求文本信息匹配度较高的商品信息，不存在现有技术中的冷启动问题，且不需要在网络模型构建的过程中人工输入商品信息，能够保证商品信息预测模型构建的整个过程自动化完成。Through the description of the foregoing implementation manners, those skilled in the art can clearly understand that this application can be implemented by means of software plus a necessary general hardware platform, or by hardware. By applying the technical solution of this application, compared with the existing technical solution for pushing product information, this embodiment can recommend and recommend the product demand text for the user based on the product demand text information input by the user through the trained product information prediction model Product information with a high degree of information matching does not have the cold start problem in the prior art, and does not require manual input of product information in the process of network model construction, which can ensure that the entire process of product information prediction model construction is automated.
本领域技术人员可以理解附图只是一个优选实施场景的示意图，附图中的模块或流程并不一定是实施本申请所必须的。本领域技术人员可以理解实施场景中的装置中的模块可以按照实施场景描述进行分布于实施场景的装置中，也可以进行相应变化位于不同于本实施场景的一个或多个装置中。上述实施场景的模块可以合并为一个模块，也可以进一步拆分成多个子模块。Those skilled in the art can understand that the accompanying drawings are only schematic diagrams of preferred implementation scenarios, and the modules or processes in the accompanying drawings are not necessarily necessary for implementing this application. Those skilled in the art can understand that the modules in the device in the implementation scenario can be distributed in the device in the implementation scenario according to the description of the implementation scenario, or can be changed to be located in one or more devices different from the implementation scenario. The modules of the above implementation scenarios can be combined into one module or further divided into multiple sub-modules.
上述本申请序号仅仅为了描述，不代表实施场景的优劣。以上公开的仅为本申请的几个具体实施场景，但是，本申请并非局限于此，任何本领域的技术人员能思之的变化都应落入本申请的保护范围。The above serial number of this application is only for description, and does not represent the merits of implementation scenarios. The above disclosures are only a few specific implementation scenarios of the application, but the application is not limited to these, and any changes that can be thought of by those skilled in the art should fall into the protection scope of the application.

Claims (20)

  1. 一种基于商品信息预测模型的商品信息推送方法，其特征在于，包括：A product information push method based on a product information prediction model, characterized in that it includes:
    根据获取到的商品需求文本信息构建数据候选集；Construct a data candidate set according to the acquired text information of commodity requirements;
    利用所构建的数据候选集对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型；Use the constructed data candidate set to train the initial product information prediction model to obtain a trained product information prediction model;
    将获取到的商品搜索请求中的商品关键词信息输入训练好的商品信息预测模型，得到与所述商品关键词信息匹配的多个商品介绍文本信息；Input the acquired product keyword information in the product search request into the trained product information prediction model to obtain multiple product introduction text information matching the product keyword information;
    分别将与所述多个商品介绍文本信息对应的多个商品信息进行显示；Respectively displaying a plurality of commodity information corresponding to the plurality of commodity introduction text information;
    其中，所述将获取到的商品搜索请求中的商品关键词信息输入训练好的商品信息预测模型，得到与所述商品关键词信息匹配的多个商品介绍文本信息，具体包括：Wherein, said inputting the acquired product keyword information in the product search request into a trained product information prediction model to obtain multiple product introduction text information matching the product keyword information specifically includes:
    利用训练好的商品信息预测模型，获取与所述商品关键词信息相匹配的多个商品信息；Use the trained product information prediction model to obtain multiple product information that matches the product keyword information;
    根据匹配得到的多个商品信息，获取与所述多个商品信息对应的多个商品介绍文本信息；According to the multiple product information obtained by matching, obtain multiple product introduction text information corresponding to the multiple product information;
    将所述商品关键词信息分别与获取到的多个商品介绍文本信息进行相似度计算，得到与所述商品关键词信息匹配的多个商品介绍文本信息。The similarity calculation is performed between the commodity keyword information and the obtained multiple commodity introduction text information to obtain multiple commodity introduction text information matching the commodity keyword information.
  2. 根据权利要求1所述的方法，其特征在于，所述商品需求文本信息包括商品关键词信息，以及对应所述商品关键词信息的多个商品介绍文本信息；The method according to claim 1, wherein the commodity demand text information includes commodity keyword information and multiple commodity introduction text information corresponding to the commodity keyword information;
    其中，用于构建数据候选集的商品需求文本信息中的对应商品关键词信息的多个商品介绍文本信息包括用户购买过的商品的第一商品介绍文本信息和用户未购买过的商品的第二商品介绍文本信息。Among them, multiple product introduction text information corresponding to the product keyword information in the product demand text information used to construct the data candidate set include the first product introduction text information of the product that the user has purchased and the second product introduction text information of the product that the user has not purchased. Product introduction text information.
  3. 根据权利要求2所述的方法，其特征在于，所述数据候选集中的第一商品介绍文本信息为正样本，所述第二商品介绍文本信息为负样本，利用所构建的数据候选集对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型，具体包括：The method according to claim 2, wherein the first product introduction text information in the data candidate set is a positive sample, the second product introduction text information is a negative sample, and the constructed data candidate set is used to initialize The commodity information prediction model is trained to obtain a trained commodity information prediction model, which specifically includes:
    将所述商品关键词信息分别与第一商品介绍文本信息和多个第二商品介绍文本信息进行距离相似度计算，得到多个相似度值；Performing distance similarity calculations on the commodity keyword information with the first commodity introduction text information and the plurality of second commodity introduction text information to obtain multiple similarity values;
    根据所述多个相似度值，计算得到所述商品关键词信息与第一商品介绍文本信息的相似度概率；According to the multiple similarity values, the similarity probability between the commodity keyword information and the first commodity introduction text information is calculated;
    根据所述相似度概率对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型。Training the initial commodity information prediction model according to the similarity probability to obtain a trained commodity information prediction model.
  4. 根据权利要求3所述的方法，其特征在于，所述根据所述相似度概率对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型，具体包括：The method according to claim 3, wherein the training the initial commodity information prediction model according to the similarity probability to obtain the trained commodity information prediction model specifically comprises:
    通过对所述相似度概率进行极大似然估计，确定商品信息预测模型的网络参数；Determine the network parameters of the commodity information prediction model by performing maximum likelihood estimation on the similarity probability;
    根据所确定的商品信息预测模型的网络参数，得到训练好的商品信息预测模型。According to the determined network parameters of the product information prediction model, a trained product information prediction model is obtained.
  5. 根据权利要求1所述的方法，其特征在于，所述利用训练好的商品信息预测模型，根据获取到的商品搜索请求中的商品关键词信息得到与所述商品关键词信息匹配的多个商品介绍文本信息之前，具体还包括：The method according to claim 1, characterized in that said using the trained commodity information prediction model, according to the obtained commodity keyword information in the commodity search request, multiple commodities matching the commodity keyword information are obtained Before introducing the text information, it also specifically includes:
    当监测到来自用户的商品搜索请求时，获取所述商品搜索请求中的商品关键词信息。When a product search request from the user is monitored, the product keyword information in the product search request is acquired.
  6. 根据权利要求1所述的方法，其特征在于，所述将所述商品关键词信息分别与获取到的多个商品介绍文本信息进行相似度计算，得到与所述商品关键词信息匹配的多个商品介绍文本信息，具体包括：The method according to claim 1, wherein the similarity calculation is performed between the product keyword information and the obtained multiple product introduction text information to obtain multiple product keyword information that match the product keyword information. Product introduction text information, including:
    将所述商品关键词信息对应的语义向量分别与获取到的多个商品介绍文本信息对应的语义向量进行距离相似度计算，得到多个相似度值；Performing distance similarity calculations on the semantic vectors corresponding to the commodity keyword information and the acquired semantic vectors corresponding to the multiple commodity introduction text information to obtain multiple similarity values;
    对计算得到的多个相似度值进行降序排列，得到降序排列结果；Sort the calculated similarity values in descending order to obtain the descending order result;
    按照预设的商品匹配值，根据所述降序排列结果得到与所述商品关键词信息匹配的多个商品介绍文本信息。According to the preset product matching value, multiple product introduction text information that matches the product keyword information are obtained according to the descending sorting result.
  7. 一种基于商品信息预测模型的商品信息推送装置，其特征在于，包括：A product information push device based on a product information prediction model, characterized in that it comprises:
    构建模块，用于根据获取到的商品需求文本信息构建数据候选集；The construction module is used to construct a data candidate set according to the acquired text information of commodity requirements;
    训练模块，用于利用所构建的数据候选集对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型；The training module is used to train the initial product information prediction model by using the constructed data candidate set to obtain the trained product information prediction model;
    匹配模块，用于将获取到的商品搜索请求中的商品关键词信息输入训练好的商品信息预测模型，得到与所述商品关键词信息匹配的多个商品介绍文本信息；The matching module is used to input the acquired product keyword information in the product search request into the trained product information prediction model to obtain multiple product introduction text information that matches the product keyword information;
    显示模块，用于分别将与所述多个商品介绍文本信息对应的多个商品信息进行显示；The display module is configured to respectively display multiple product information corresponding to the multiple product introduction text information;
    其中，所述匹配模块，具体包括：Wherein, the matching module specifically includes:
    第一获取单元，用于利用训练好的商品信息预测模型，获取与所述商品关键词信息相匹配的多个商品信息；The first obtaining unit is configured to use the trained commodity information prediction model to obtain multiple commodity information matching the commodity keyword information;
    第二获取单元，用于根据匹配得到的多个商品信息，获取与所述多个商品信息对应的多个商品介绍文本信息；The second acquiring unit is configured to acquire multiple product introduction text information corresponding to the multiple product information according to multiple product information obtained by matching;
    相似计算单元，用于将所述商品关键词信息分别与获取到的多个商品介绍文本信息进行相似度计算，得到与所述商品关键词信息匹配的多个商品介绍文本信息。The similarity calculation unit is used to calculate the similarity between the product keyword information and the obtained multiple product introduction text information to obtain multiple product introduction text information matching the product keyword information.
  8. 根据权利要求7所述的装置，其特征在于，所述商品需求文本信息包括商品关键词信息，以及对应所述商品关键词信息的多个商品介绍文本信息；8. The device according to claim 7, wherein the commodity demand text information includes commodity keyword information and multiple commodity introduction text information corresponding to the commodity keyword information;
    其中，用于构建数据候选集的商品需求文本信息中的对应商品关键词信息的多个商品介绍文本信息包括用户购买过的商品的第一商品介绍文本信息和用户未购买过的商品的第二商品介绍文本信息。Among them, multiple product introduction text information corresponding to the product keyword information in the product demand text information used to construct the data candidate set include the first product introduction text information of the product that the user has purchased and the second product introduction text information of the product that the user has not purchased. Product introduction text information.
  9. 根据权利要求8所述的装置，其特征在于，所述数据候选集中的第一商品介绍文本信息为正样本，所述第二商品介绍文本信息为负样本，所述训练模块，具体包括：The device according to claim 8, wherein the first product introduction text information in the data candidate set is a positive sample, and the second product introduction text information is a negative sample, and the training module specifically includes:
    第一计算单元，用于将所述商品关键词信息分别与第一商品介绍文本信息和多个第二商品介绍文本信息进行距离相似度计算，得到多个相似度值；The first calculation unit is configured to calculate the distance similarity between the commodity keyword information and the first commodity introduction text information and multiple second commodity introduction text information respectively to obtain multiple similarity values;
    第二计算单元，用于根据所述多个相似度值，计算得到所述商品关键词信息与第一商品介绍文本信息的相似度概率；The second calculation unit is configured to calculate the similarity probability between the commodity keyword information and the first commodity introduction text information according to the multiple similarity values;
    最小化训练单元，用于根据所述相似度概率对初始化商品信息预测模型进行训练，得到训练好的商品 信息预测模型。The minimization training unit is used to train the initial product information prediction model according to the similarity probability to obtain the trained product information prediction model.
  10. 根据权利要求9所述的装置，其特征在于，所述最小化训练单元，具体包括：The device according to claim 9, wherein the minimization training unit specifically comprises:
    通过对所述相似度概率进行极大似然估计，确定商品信息预测模型的网络参数；Determine the network parameters of the commodity information prediction model by performing maximum likelihood estimation on the similarity probability;
    根据所确定的商品信息预测模型的网络参数，得到训练好的商品信息预测模型。According to the determined network parameters of the product information prediction model, a trained product information prediction model is obtained.
  11. 根据权利要求7所述的装置，其特征在于，还包括监测模块，具体包括：The device according to claim 7, characterized in that it further comprises a monitoring module, specifically comprising:
    当监测到来自用户的商品搜索请求时，获取所述商品搜索请求中的商品关键词信息。When a product search request from the user is monitored, the product keyword information in the product search request is acquired.
  12. 根据权利要求7所述的装置，其特征在于，所述相似计算单元，具体包括：The device according to claim 7, wherein the similarity calculation unit specifically comprises:
    将所述商品关键词信息对应的语义向量分别与获取到的多个商品介绍文本信息对应的语义向量进行距离相似度计算，得到多个相似度值；Performing distance similarity calculations on the semantic vectors corresponding to the commodity keyword information and the acquired semantic vectors corresponding to the multiple commodity introduction text information to obtain multiple similarity values;
    对计算得到的多个相似度值进行降序排列，得到降序排列结果；Sort the calculated similarity values in descending order to obtain the descending order result;
    按照预设的商品匹配值，根据所述降序排列结果得到与所述商品关键词信息匹配的多个商品介绍文本信息。According to the preset product matching value, multiple product introduction text information that matches the product keyword information are obtained according to the descending sorting result.
  13. 一种非易失性可读存储介质，其上存储有计算机可读指令，其特征在于，所述程序被处理器执行时实现基于商品信息预测模型的商品信息推送方法，包括：A non-volatile readable storage medium having computer readable instructions stored thereon, characterized in that, when the program is executed by a processor, a method for pushing product information based on a product information prediction model is realized, including:
    根据获取到的商品需求文本信息构建数据候选集；Construct a data candidate set according to the acquired text information of commodity requirements;
    利用所构建的数据候选集对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型；Use the constructed data candidate set to train the initial product information prediction model to obtain a trained product information prediction model;
    将获取到的商品搜索请求中的商品关键词信息输入训练好的商品信息预测模型，得到与所述商品关键词信息匹配的多个商品介绍文本信息；Input the acquired product keyword information in the product search request into the trained product information prediction model to obtain multiple product introduction text information matching the product keyword information;
    分别将与所述多个商品介绍文本信息对应的多个商品信息进行显示；Respectively displaying a plurality of commodity information corresponding to the plurality of commodity introduction text information;
    其中，所述将获取到的商品搜索请求中的商品关键词信息输入训练好的商品信息预测模型，得到与所述商品关键词信息匹配的多个商品介绍文本信息，具体包括：Wherein, said inputting the acquired product keyword information in the product search request into a trained product information prediction model to obtain multiple product introduction text information matching the product keyword information specifically includes:
    利用训练好的商品信息预测模型，获取与所述商品关键词信息相匹配的多个商品信息；Use the trained product information prediction model to obtain multiple product information that matches the product keyword information;
    根据匹配得到的多个商品信息，获取与所述多个商品信息对应的多个商品介绍文本信息；According to the multiple product information obtained by matching, obtain multiple product introduction text information corresponding to the multiple product information;
    将所述商品关键词信息分别与获取到的多个商品介绍文本信息进行相似度计算，得到与所述商品关键词信息匹配的多个商品介绍文本信息。The similarity calculation is performed between the commodity keyword information and the obtained multiple commodity introduction text information to obtain multiple commodity introduction text information matching the commodity keyword information.
  14. 根据权利要求13所述的非易失性可读存储介质，其特征在于，所述商品需求文本信息包括商品关键词信息，以及对应所述商品关键词信息的多个商品介绍文本信息；The non-volatile readable storage medium according to claim 13, wherein the commodity demand text information includes commodity keyword information and multiple commodity introduction text information corresponding to the commodity keyword information;
    其中，用于构建数据候选集的商品需求文本信息中的对应商品关键词信息的多个商品介绍文本信息包括用户购买过的商品的第一商品介绍文本信息和用户未购买过的商品的第二商品介绍文本信息。Among them, multiple product introduction text information corresponding to the product keyword information in the product demand text information used to construct the data candidate set include the first product introduction text information of the product that the user has purchased and the second product introduction text information of the product that the user has not purchased. Product introduction text information.
  15. 根据权利要求14所述的非易失性可读存储介质，其特征在于，所述数据候选集中的第一商品介绍文本信息为正样本，所述第二商品介绍文本信息为负样本，利用所构建的数据候选集对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型，具体包括：The non-volatile readable storage medium according to claim 14, wherein the first product introduction text information in the data candidate set is a positive sample, and the second product introduction text information is a negative sample. The constructed data candidate set trains the initial product information prediction model to obtain a trained product information prediction model, which specifically includes:
    将所述商品关键词信息分别与第一商品介绍文本信息和多个第二商品介绍文本信息进行距离相似度计算，得到多个相似度值；Performing distance similarity calculations on the commodity keyword information with the first commodity introduction text information and the plurality of second commodity introduction text information to obtain multiple similarity values;
    根据所述多个相似度值，计算得到所述商品关键词信息与第一商品介绍文本信息的相似度概率；According to the multiple similarity values, the similarity probability between the commodity keyword information and the first commodity introduction text information is calculated;
    根据所述相似度概率对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型。Training the initial commodity information prediction model according to the similarity probability to obtain a trained commodity information prediction model.
  16. 根据权利要求15所述的非易失性可读存储介质，其特征在于，所述根据所述相似度概率对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型，具体包括：The non-volatile readable storage medium according to claim 15, wherein the training of the initial commodity information prediction model according to the similarity probability to obtain the trained commodity information prediction model specifically comprises:
    通过对所述相似度概率进行极大似然估计，确定商品信息预测模型的网络参数；Determine the network parameters of the commodity information prediction model by performing maximum likelihood estimation on the similarity probability;
    根据所确定的商品信息预测模型的网络参数，得到训练好的商品信息预测模型。According to the determined network parameters of the product information prediction model, a trained product information prediction model is obtained.
  17. 一种计算机设备，包括非易失性可读存储介质、处理器及存储在非易失性可读存储介质上并可在处理器上运行的计算机可读指令，其特征在于，所述处理器执行所述程序时实现基于商品信息预测模型的商品信息推送方法，包括：A computer device, including a non-volatile readable storage medium, a processor, and computer readable instructions stored on the non-volatile readable storage medium and running on the processor, characterized in that the processor The method for pushing commodity information based on the commodity information prediction model when the program is executed includes:
    根据获取到的商品需求文本信息构建数据候选集；Construct a data candidate set according to the acquired text information of commodity requirements;
    利用所构建的数据候选集对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型；Use the constructed data candidate set to train the initial product information prediction model to obtain a trained product information prediction model;
    将获取到的商品搜索请求中的商品关键词信息输入训练好的商品信息预测模型，得到与所述商品关键词信息匹配的多个商品介绍文本信息；Input the acquired product keyword information in the product search request into the trained product information prediction model to obtain multiple product introduction text information matching the product keyword information;
    分别将与所述多个商品介绍文本信息对应的多个商品信息进行显示；Respectively displaying a plurality of commodity information corresponding to the plurality of commodity introduction text information;
    其中，所述将获取到的商品搜索请求中的商品关键词信息输入训练好的商品信息预测模型，得到与所述商品关键词信息匹配的多个商品介绍文本信息，具体包括：Wherein, said inputting the acquired product keyword information in the product search request into a trained product information prediction model to obtain multiple product introduction text information matching the product keyword information specifically includes:
    利用训练好的商品信息预测模型，获取与所述商品关键词信息相匹配的多个商品信息；Use the trained product information prediction model to obtain multiple product information that matches the product keyword information;
    根据匹配得到的多个商品信息，获取与所述多个商品信息对应的多个商品介绍文本信息；According to the multiple product information obtained by matching, obtain multiple product introduction text information corresponding to the multiple product information;
    将所述商品关键词信息分别与获取到的多个商品介绍文本信息进行相似度计算，得到与所述商品关键词信息匹配的多个商品介绍文本信息。The similarity calculation is performed between the commodity keyword information and the obtained multiple commodity introduction text information to obtain multiple commodity introduction text information matching the commodity keyword information.
  18. 根据权利要求17所述的计算机设备，其特征在于，所述商品需求文本信息包括商品关键词信息，以及对应所述商品关键词信息的多个商品介绍文本信息；The computer device according to claim 17, wherein the commodity demand text information includes commodity keyword information, and multiple commodity introduction text information corresponding to the commodity keyword information;
    其中，用于构建数据候选集的商品需求文本信息中的对应商品关键词信息的多个商品介绍文本信息包括用户购买过的商品的第一商品介绍文本信息和用户未购买过的商品的第二商品介绍文本信息。Among them, multiple product introduction text information corresponding to the product keyword information in the product demand text information used to construct the data candidate set include the first product introduction text information of the product that the user has purchased and the second product introduction text information of the product that the user has not purchased. Product introduction text information.
  19. 根据权利要求18所述的计算机设备，其特征在于，所述数据候选集中的第一商品介绍文本信息为正样本，所述第二商品介绍文本信息为负样本，利用所构建的数据候选集对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型，具体包括：The computer device according to claim 18, wherein the first product introduction text information in the data candidate set is a positive sample, the second product introduction text information is a negative sample, and the constructed data candidate set is used to pair Initialize the product information prediction model for training, and obtain a trained product information prediction model, including:
    将所述商品关键词信息分别与第一商品介绍文本信息和多个第二商品介绍文本信息进行距离相似度计算，得到多个相似度值；Performing distance similarity calculations on the commodity keyword information with the first commodity introduction text information and the plurality of second commodity introduction text information to obtain multiple similarity values;
    根据所述多个相似度值，计算得到所述商品关键词信息与第一商品介绍文本信息的相似度概率；According to the multiple similarity values, the similarity probability between the commodity keyword information and the first commodity introduction text information is calculated;
    根据所述相似度概率对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型。Training the initial commodity information prediction model according to the similarity probability to obtain a trained commodity information prediction model.
  20. 根据权利要求19所述的计算机设备，其特征在于，所述根据所述相似度概率对初始化商品信息预测模型进行训练，得到训练好的商品信息预测模型，具体包括：The computer device according to claim 19, wherein the training of the initial commodity information prediction model according to the similarity probability to obtain the trained commodity information prediction model specifically comprises:
    通过对所述相似度概率进行极大似然估计，确定商品信息预测模型的网络参数；Determine the network parameters of the commodity information prediction model by performing maximum likelihood estimation on the similarity probability;
    根据所确定的商品信息预测模型的网络参数，得到训练好的商品信息预测模型。According to the determined network parameters of the product information prediction model, a trained product information prediction model is obtained.
PCT/CN2019/118488 2019-08-22 2019-11-14 Commodity information prediction model-based commodity information pushing method and apparatus, non-volatile readable storage medium, and computer device WO2021031410A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910780785.1 2019-08-22
CN201910780785.1A CN110706055A (en) 2019-08-22 2019-08-22 Commodity information pushing method and device, storage medium and computer equipment

Publications (1)

Publication Number Publication Date
WO2021031410A1 true WO2021031410A1 (en) 2021-02-25

Family

ID=69193799

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/118488 WO2021031410A1 (en) 2019-08-22 2019-11-14 Commodity information prediction model-based commodity information pushing method and apparatus, non-volatile readable storage medium, and computer device

Country Status (2)

Country Link
CN (1) CN110706055A (en)
WO (1) WO2021031410A1 (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113450172A (en) * 2020-03-27 2021-09-28 北京沃东天骏信息技术有限公司 Commodity recommendation method and device
CN111611807B (en) * 2020-05-18 2022-12-09 北京邮电大学 Keyword extraction method and device based on neural network and electronic equipment
CN111681086A (en) * 2020-06-16 2020-09-18 上海风秩科技有限公司 Commodity recommendation method and device, computer equipment and readable storage medium
CN111737418B (en) * 2020-07-20 2021-05-14 北京每日优鲜电子商务有限公司 Method, apparatus and storage medium for predicting relevance of search term and commodity
CN115250365A (en) * 2021-04-28 2022-10-28 京东科技控股股份有限公司 Commodity text generation method and device, computer equipment and storage medium

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105469263A (en) * 2014-09-24 2016-04-06 阿里巴巴集团控股有限公司 Commodity recommendation method and device
US20160125501A1 (en) * 2014-11-04 2016-05-05 Philippe Nemery Preference-elicitation framework for real-time personalized recommendation
CN106295832A (en) * 2015-05-12 2017-01-04 阿里巴巴集团控股有限公司 Product information method for pushing and device
CN108711110A (en) * 2018-08-14 2018-10-26 中国平安人寿保险股份有限公司 Insurance products recommend method, apparatus, computer equipment and storage medium
CN108960945A (en) * 2017-05-18 2018-12-07 北京京东尚科信息技术有限公司 Method of Commodity Recommendation and device
CN109711929A (en) * 2018-12-13 2019-05-03 中国平安财产保险股份有限公司 Business recommended method and device based on prediction model

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109344314B (en) * 2018-08-20 2021-11-16 腾讯科技（深圳）有限公司 Data processing method and device and server

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105469263A (en) * 2014-09-24 2016-04-06 阿里巴巴集团控股有限公司 Commodity recommendation method and device
US20160125501A1 (en) * 2014-11-04 2016-05-05 Philippe Nemery Preference-elicitation framework for real-time personalized recommendation
CN106295832A (en) * 2015-05-12 2017-01-04 阿里巴巴集团控股有限公司 Product information method for pushing and device
CN108960945A (en) * 2017-05-18 2018-12-07 北京京东尚科信息技术有限公司 Method of Commodity Recommendation and device
CN108711110A (en) * 2018-08-14 2018-10-26 中国平安人寿保险股份有限公司 Insurance products recommend method, apparatus, computer equipment and storage medium
CN109711929A (en) * 2018-12-13 2019-05-03 中国平安财产保险股份有限公司 Business recommended method and device based on prediction model

Also Published As

Publication number Publication date
CN110706055A (en) 2020-01-17

Similar Documents

Publication Publication Date Title
WO2021031410A1 (en) Commodity information prediction model-based commodity information pushing method and apparatus, non-volatile readable storage medium, and computer device
CN110472090B (en) Image retrieval method based on semantic tags, related device and storage medium
WO2020125445A1 (en) Classification model training method, classification method, device and medium
WO2019137104A1 (en) Recommendation method and device employing deep learning, electronic apparatus, medium, and program
WO2017181612A1 (en) Personalized video recommendation method and device
WO2018192491A1 (en) Information pushing method and device
WO2018040944A1 (en) System, method, and device for identifying malicious address/malicious purchase order
CN108280115B (en) Method and device for identifying user relationship
WO2017000109A1 (en) Search method, search apparatus, user equipment, and computer program product
CN112364204B (en) Video searching method, device, computer equipment and storage medium
US20220084102A1 (en) Commodity recommendation method, server, shopping cart and shopping system
US20220417339A1 (en) Feature-based network embedding
WO2020007177A1 (en) Quotation method executed by computer, quotation device, electronic device and storage medium
WO2023011382A1 (en) Recommendation method, recommendation model training method, and related product
WO2020119284A1 (en) Method and device for determining risk of user access
WO2019072098A1 (en) Method and system for identifying core product terms
US20150235264A1 (en) Automatic entity detection and presentation of related content
CN111782765A (en) Recommendation method based on graph attention machine mechanism
WO2023168810A1 (en) Method and apparatus for predicting properties of drug molecule, storage medium, and computer device
CN110569446B (en) Method and system for constructing recommended object candidate set
CN111626767A (en) Resource data distribution method, device and equipment
US20230245210A1 (en) Knowledge graph-based information recommendation
WO2020098163A1 (en) Product data pushing method and device, and computer non-volatile readable storage medium
CN113435523B (en) Method, device, electronic equipment and storage medium for predicting content click rate
WO2019056502A1 (en) Variety game result prediction method and apparatus, and storage medium

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19942498

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19942498

Country of ref document: EP

Kind code of ref document: A1