WO2023035941A1 - 一种承揽服务推荐方法及系统 - Google Patents

一种承揽服务推荐方法及系统 Download PDF

Info

Publication number
WO2023035941A1
WO2023035941A1 PCT/CN2022/114377 CN2022114377W WO2023035941A1 WO 2023035941 A1 WO2023035941 A1 WO 2023035941A1 CN 2022114377 W CN2022114377 W CN 2022114377W WO 2023035941 A1 WO2023035941 A1 WO 2023035941A1
Authority
WO
WIPO (PCT)
Prior art keywords
label
contract
target
processed
neural network
Prior art date
Application number
PCT/CN2022/114377
Other languages
English (en)
French (fr)
Inventor
姚娟娟
钟南山
Original Assignee
上海明品医学数据科技有限公司
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 上海明品医学数据科技有限公司 filed Critical 上海明品医学数据科技有限公司
Publication of WO2023035941A1 publication Critical patent/WO2023035941A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35—Clustering; Classification
    • G06F16/355—Class or cluster creation or modification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/335—Filtering based on additional data, e.g. user or group profiles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/958—Organisation or management of web site content, e.g. publishing, maintaining pages or automatic linking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Definitions

  • the present invention relates to the field of big data technology, in particular to a contracting service recommendation method and system.
  • the contracting service refers to the behavior that the party/contractor completes a certain work for the target party, and the target party pays the agreed remuneration after acceptance.
  • the target object In contracting services, the target object is often unable to clarify or state its own needs, and it is also difficult to choose a matching undertaking object according to its own needs. Therefore, the target object often pays a large opportunity cost.
  • the purpose of the present invention is to provide a contracting service recommendation method and system for solving the problem of inaccurate contracting service recommendation in the prior art.
  • an embodiment of the present invention provides a method for recommending contracted services, including:
  • the contractor of the target object to be measured is determined according to the predicted confidence.
  • the step of the first classification process includes: inputting the target text into the first neural network to obtain a processing result, and comparing the similarity between the processing result and the labeled target label, and according to the similarity Obtain the processed target label;
  • the step of the second classification process includes: inputting the contract text into the second neural network to obtain a processing result, comparing the processing result with the labeled contract label, and obtaining the processed contract according to the similarity. Label.
  • the third neural network includes an input layer, a hidden layer, and an output layer, and the third neural network is trained through a loss function, and the loss function includes a first loss function, a second loss function, and a third loss function, the first loss function includes the loss between the processed target label and the marked target label, and the second loss function includes the difference between the processed contract label and the marked contract label.
  • the loss between the third loss function includes the loss between the processed target label and the predicted contract label.
  • L is a loss function
  • L1 is a first loss function
  • L2 is a second loss function
  • L3 is a third loss function
  • N is the number of the processed target labels
  • M is the number of the processed contract labels quantity
  • pie is the probability that the i-th label in the set of processed target labels matches the e-th label in the set of marked target labels, and pjf is the match between the j-th label in the set of processed contract labels and The probability that the f-th label in the set of labeled contract labels matches;
  • pk is the probability that the processed target label matches the predicted contract label.
  • the expected parameter is determined through the predicted confidence
  • the contracting object of the target object to be measured is determined through the expected parameter
  • the mathematical expression of the expected parameter is:
  • Q is the expected parameter
  • C is the quantity of the target label to be tested
  • a is the ath target label to be tested
  • D is the quantity of the predicted contract label
  • b is the bth predicted Contract label
  • Pa(Kb) is the confidence degree of the b-th predicted contract label corresponding to the a-th target label to be tested.
  • the expected value is determined through the expected parameter
  • the contracted object of the target object to be measured is determined through the ranking of the expected value of each of the contracted objects, and the mathematical expression of the expected value is:
  • W is the expected value
  • H is a constant greater than 1
  • Q is the expected parameter
  • the activation function of the output layer is a sigmod function.
  • a contract service recommendation system comprising:
  • An acquisition module configured to acquire the target text of the target object and the contract text of the contract object
  • a model module configured to input the target text into a first neural network for first classification processing, obtain a processed target label, and input the contract text into a second neural network for second classification processing, to obtain For the processed contract label, perform fusion processing on the processed target label and the processed contract label to obtain a fusion label, and input the fusion label into a third neural network for a third classification process to obtain Contracting tags and confidence levels matched with the processed target tags, and obtaining a classification model by training the third neural network;
  • the processing module is used to fuse the target label of the target object to be tested with the contract label of the contract object to be tested, obtain the fusion label to be tested and input it into the classification model, and obtain the predicted contract label and the predicted confidence Spend;
  • a matching module configured to determine the contractor of the target object to be tested according to the predicted confidence.
  • An electronic device comprising:
  • One or more processors and one or more machine-readable media having instructions stored thereon, which, when executed by the one or more processors, cause the electronic device to perform any one of the contract service recommendation methods .
  • a machine-readable medium stores instructions thereon, which, when executed by one or more processors, cause a device to execute the method for recommending contracted services.
  • FIG. 1 is a schematic diagram of a contracted service recommendation method according to an embodiment of the present invention.
  • FIG. 2 is a schematic diagram of a contracted service recommendation system according to an embodiment of the present invention.
  • one embodiment of the present invention provides a contracted service recommendation method, including:
  • S1 Obtain the target text of the target object and the contract text of the contract object.
  • the information carried by the target text is associated with the needs of the target object.
  • the information of the contract text is related to the contract object provided by the contract object.
  • the contracting service when the needs of the target object match the contracting service of the contracting object, it can not only meet the needs of the target object, but also facilitate the contracting service provided by the contracting object.
  • S2 Input the target text into the first neural network for the first classification process, obtain the processed target label, and input the contract text into the second neural network for the second classification process, and obtain the processed target label
  • the contract label, the target label or the contract label is a vector, the processed target label and the processed contract label are fused to obtain a fusion label, the fusion label is a feature matrix, and the The fusion label is input into the third neural network for the third classification process, and the contract label and confidence degree matched with the processed target label are obtained, and the classification model is obtained by training the third neural network, for example, by Iterative training, detecting the accuracy rate (Precision) and recall rate (Recall) of matching the target label and the contract label, and for example, detecting the F value (F-Measure) of the confidence level, and obtaining the preferred training model as a classification model;
  • Precision accuracy rate
  • Recall recall rate
  • S3 Fuse the target label of the target object to be tested with the contract label of the contract object to be tested, obtain the fusion label to be tested and input it into the classification model, obtain the predicted contract label and the predicted confidence, and obtain
  • the steps of the target label and the contract label include: classifying the text through the neural network to obtain the label, for example, obtaining the target label through the first neural network process, and for example, obtaining the contract label through the second neural network process;
  • S4 Determine the contracting object of the target object to be measured according to the confidence level of the prediction, therefore, the contracting object can provide better contracting services for the target object.
  • the relevant information of the target object and the contracting object is obtained, and the classification logic of the target object and the contracting object is learned through machine learning, and the classification model with the classification logic is obtained through model training.
  • the target to be tested Objects are assigned to matching contract objects through the processing of the classification model.
  • the step of the first classification processing includes: inputting the target text into the first neural network to obtain a processing result, and comparing the similarity between the processing result and the marked target label, Obtain the processed target label according to the similarity;
  • the step of the second classification process includes: inputting the contract text into the second neural network to obtain a processing result, comparing the processing result with the labeled contract label, and obtaining the processed contract according to the similarity. Label.
  • the label information is obtained from the text information.
  • the target label or contract label can be vectorized, for example, the target vector can be obtained through the target label, and for example, the contract can be used The label obtains the contract vector, which facilitates the fusion of the target vector and the contract vector.
  • the third neural network includes an input layer, a hidden layer, and an output layer, and the third neural network is trained through a loss function, and the loss function includes: a first loss function, a second loss function, and a third A loss function, the first loss function includes the loss between the processed target label and the marked target label, and the second loss function includes the processed contract label and the marked contract label.
  • the loss between the third loss function includes the loss between the processed target label and the predicted contract label.
  • the mathematical expression of the loss function is:
  • L is a loss function
  • L1 is a first loss function
  • L2 is a second loss function
  • L3 is a third loss function
  • N is the number of the processed target labels
  • M is the number of the processed contract labels quantity
  • pie is the probability that the i-th label in the set of processed target labels matches the e-th label in the set of marked target labels, and pjf is the match between the j-th label in the set of processed contract labels and The probability that the f-th label in the set of labeled contract labels matches;
  • pk is the probability that the processed target label matches the predicted contract label.
  • This loss function not only considers the loss between the processed target label and the marked target label, the loss between the processed contract label and the marked contract label, but also considers the processed The loss between the target label and the predicted take label.
  • the third neural network and data are trained through the loss function to obtain an ideal classification model, which improves the classification accuracy of the classification model, and obtains classification logic through the classification model.
  • the expected parameter is determined through the confidence of the prediction, and the contracting object of the target object to be measured is determined through the expected parameter, and the mathematical expression of the expected parameter is:
  • Q is the expected parameter
  • C is the number of the target tags to be processed
  • a is the a-th target tag to be processed
  • D is the number of the predicted contract tags
  • b is the b-th predicted Contract label
  • Pa(Kb) is the confidence degree of the b-th predicted contract label corresponding to the a-th target label to be processed.
  • the target object to be tested has C target labels
  • there are multiple contract objects to be tested and the contract object has D contract labels
  • the weighted value of the confidence of the D contract labels is calculated, and then the contract object to be tested is measured against
  • the size of the expected parameter is used as an index to measure the matching relationship.
  • the above-mentioned matching relationship can also be measured by the expected value, for example, the expected value is determined through the expected parameter, and the contracted object of the target object to be measured is determined through the ranking of the expected value of each of the contracted objects, so
  • the mathematical expression of the expected value is:
  • W is the expected value
  • H is a constant greater than 1
  • Q is the expected parameter
  • the activation function of the output layer can be set as a sigmod function.
  • the present invention provides a contracted service recommendation system, including:
  • An acquisition module configured to acquire the target text of the target object and the contract text of the contract object
  • a model module configured to input the target text into a first neural network for first classification processing, obtain a processed target label, and input the contract text into a second neural network for second classification processing, to obtain For the processed contract label, perform fusion processing on the processed target label and the processed contract label to obtain a fusion label, and input the fusion label into a third neural network for a third classification process to obtain Contracting tags and confidence levels matched with the processed target tags, and obtaining a classification model by training a third neural network;
  • the processing module is used to fuse the target label of the target object to be tested with the contract label of the contract object to be tested, obtain the fusion label to be tested and input it into the classification model, and obtain the predicted contract label and the predicted confidence Spend;
  • a matching module configured to determine the contractor of the target object to be tested according to the predicted confidence.
  • the step of the first classification process includes: inputting the target text into the first neural network to obtain a processing result, and comparing the similarity between the processing result and the labeled target label, and according to the similarity Obtain the processed target label;
  • the step of the second classification process includes: inputting the contract text into the second neural network to obtain a processing result, comparing the processing result with the labeled contract label, and obtaining the processed contract according to the similarity. Label.
  • the third neural network includes an input layer, a hidden layer, and an output layer, and the third neural network is trained through a loss function, and the loss function includes a first loss function, a second loss function, and a third loss function, the first loss function includes the loss between the processed target label and the marked target label, and the second loss function includes the difference between the processed contract label and the marked contract label.
  • the loss between the third loss function includes the loss between the processed target label and the predicted contract label.
  • L is a loss function
  • L1 is a first loss function
  • L2 is a second loss function
  • L3 is a third loss function
  • N is the number of the processed target labels
  • M is the number of the processed contract labels quantity
  • pie is the probability that the i-th label in the set of processed target labels matches the e-th label in the set of marked target labels, and pjf is the match between the j-th label in the set of processed contract labels and The probability that the f-th label in the set of labeled contract labels matches;
  • pk is the probability that the processed target label matches the predicted contract label.
  • the expected parameter is determined through the predicted confidence
  • the contracting object of the target object to be measured is determined through the expected parameter
  • the mathematical expression of the expected parameter is:
  • Q is the expected parameter
  • C is the number of the target tags to be tested
  • a is the ath target tag to be tested
  • D is the number of the predicted contract tags
  • b is the bth predicted Contract label
  • Pa(Kb) is the confidence degree of the b-th predicted contract label corresponding to the a-th target label to be tested.
  • the expected value is determined through the expected parameter
  • the contracted object of the target object to be measured is determined through the ranking of the expected value of each of the contracted objects, and the mathematical expression of the expected value is:
  • W is the expected value
  • H is a constant greater than 1
  • Q is the expected parameter
  • the activation function of the output layer is a sigmod function.
  • An embodiment of the present invention provides an electronic device, including: one or more processors; and one or more machine-readable media storing instructions thereon, and when the one or more processors are executed, the An electronic device performs one or more of the described methods.
  • the invention is applicable to numerous general purpose and special purpose computing system environments or configurations. Examples: personal computers, server computers, handheld or portable devices, tablet-type devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, including A distributed computing environment for any of the above systems or devices, etc.
  • the embodiments of the present invention also provide one or more machine-readable media, on which are stored instructions, which, when executed by one or more processors, cause the device to perform one or more of the above-mentioned methods.
  • the invention may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer.
  • program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types.
  • the invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network.
  • program modules may be located in both local and remote computer storage media including storage devices.
  • the contracted service recommendation method and system of the present invention have the following beneficial effects:
  • the relevant information of the target object and the contracting object is obtained, and the classification logic of the target object and the contracting object is learned through machine learning, and the classification model with the classification logic is obtained through model training.
  • the target to be tested Objects are assigned to matching contract objects through the processing of the classification model.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Biology (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Medical Informatics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

本发明提供一种承揽服务推荐方法及系统，承揽服务推荐方法包括：获取目标对象的目标文本和承揽对象的承揽文本；将所述目标文本输入到第一神经网络中进行第一分类处理，获取处理后的目标标签，并将所述承揽文本输入到第二神经网络中进行第二分类处理，获得处理后的承揽标签，将处理后的所述目标标签和处理后的所述承揽标签进行融合处理，获取融合标签，并将所述融合标签输入到第三神经网络中进行第三分类处理，获取与处理后的所述目标标签相匹配的承揽标签以及置信度，并通过训练获取分类模型；将待测的目标对象的目标标签输入到所述分类模型中，获取预测的承揽标签以及预测的置信度；根据所述预测的置信度确定所述待测的目标对象的承揽对象。

Description

一种承揽服务推荐方法及系统 技术领域
本发明涉及大数据技术领域，特别是涉及一种承揽服务推荐方法及系统。
背景技术
随着经济社会的发展，普通大众对于服务的需求越来越专业化和细分化，受限制于获得服务的方式和途径，需求方往往均不能找到精确的、理想的服务方，尤其体现在专业性较强的承揽服务中，其中，承揽服务是指当事人/承揽对象为目标对象一方完成一定的工作，目标对象在验收后支付约定的报酬的行为。
在承揽服务中，目标对象往往不能明确或者陈述自身的需求，也难以按照自身需求选择匹配的承担对象，因此，目标对象往往付出了较大的机会成本。
发明内容
鉴于以上所述现有技术的缺点，本发明的目的在于提供一种承揽服务推荐方法及系统，用于解决现有技术中承揽服务推荐不精确的问题。
为实现上述目的及其他相关目的，本发明的实施例提供一种承揽服务推荐方法，包括：
获取目标对象的目标文本和承揽对象的承揽文本；
将所述目标文本输入到第一神经网络中进行第一分类处理，获取处理后的目标标签，并将所述承揽文本输入到第二神经网络中进行第二分类处理，获得处理后的承揽标签，将处理后的所述目标标签和处理后的所述承揽标签进行融合处理，获取融合标签，并将所述融合标签输入到第三神经网络中进行第三分类处理，获取与处理后的所述目标标签相匹配的承揽标签以及置信度，并通过训练所述第三神经网络获取分类模型；
将待测的目标对象的目标标签和待测的承揽对象的承揽标签进行融合，得到待测的融合标签并输入到所述分类模型中，获取预测的承揽标签以及预测的置信度；
根据所述预测的置信度确定所述待测的目标对象的承揽对象。
进一步地，所述第一分类处理的步骤包括：将所述目标文本输入到所述第一神经网络中获取处理结果，并将所述处理结果与标注的目标标签进行相似度对比，根据相似度获取处理后的目标标签；
所述第二分类处理的步骤包括：将所述承揽文本输入到所述第二神经网络中获取处理结果，并将处理结果与标注的承揽标签进行相似度对比，根据相似度获取处理后的承揽标签。
进一步地，所述第三神经网络包括输入层、隐藏层以及输出层，通过损失函数对所述第三神经网络进行训练，所述损失函数包括第一损失函数、第二损失函数以及第三损失函数， 所述第一损失函数包括所述处理后的目标标签与所述标注的目标标签之间的损失，所述第二损失函数包括所述处理后的承揽标签与所述标注的承揽标签之间的损失，所述第三损失函数包括所述处理后的目标标签与所述预测的承揽标签之间的损失。
进一步地，所述损失函数的数学表达为：
L＝L1+L2+L3
Figure PCTCN2022114377-appb-000001
Figure PCTCN2022114377-appb-000002
Figure PCTCN2022114377-appb-000003
其中，L为损失函数，L1为第一损失函数，L2为第二损失函数，L3为第三损失函数，N为所述处理后的目标标签的数量，M为所述处理后的承揽标签的数量，
当所述处理后的目标标签的集合中第i个标签与标注的目标标签的集合中的第e个标签匹配时，x(ie)＝1，否则，x(ie)＝0；
当所述处理后的承揽标签的集合中第j个标签与标注的承揽标签的集合中的第f个标签匹配时，y(jf)＝1，否则，y(jf)＝0；
pie为所述处理后的目标标签的集合中第i个标签与标注的目标标签的集合中的第e个标签匹配的概率，pjf为所述处理后的承揽标签的集合中第j个标签与标注的承揽标签的集合中的第f个标签匹配的概率；
当处理后的目标标签与所述预测的承揽标签匹配时，zk＝1，否则，zk＝0；
pk为处理后的目标标签与所述预测的承揽标签匹配的概率。
进一步地，通过所述预测的置信度确定期望参数，并通过所述期望参数确定所述待测的目标对象的承揽对象，所述期望参数的数学表达为：
Figure PCTCN2022114377-appb-000004
其中，Q为所述期望参数，C为所述待测的目标标签的数量，a为第a个待测的目标标签， D为所述预测的承揽标签的数量，b为第b个预测的承揽标签，Pa(Kb)为第a个待测的目标标签所对应的第b个预测的承揽标签的置信度。
进一步地，通过所述期望参数确定期望值，并通过各个所述承揽对象的所述期望值的排名，确定所述待测的目标对象的承揽对象，所述期望值的数学表达为：
W＝H Q
其中，W为所述期望值，H为大于1的常数，Q为所述期望参数。
进一步地，所述输出层的激活函数为sigmod函数。
一种承揽服务推荐系统，包括：
获取模块，用于获取目标对象的目标文本和承揽对象的承揽文本；
模型模块，用于将所述目标文本输入到第一神经网络中进行第一分类处理，获取处理后的目标标签，并将所述承揽文本输入到第二神经网络中进行第二分类处理，获得处理后的承揽标签，将处理后的所述目标标签和处理后的所述承揽标签进行融合处理，获取融合标签，并将所述融合标签输入到第三神经网络中进行第三分类处理，获取与处理后的所述目标标签相匹配的承揽标签以及置信度，并通过训练所述第三神经网络获取分类模型；
处理模块，用于将待测的目标对象的目标标签和待测的承揽对象的承揽标签进行融合，得到待测的融合标签并输入到所述分类模型中，获取预测的承揽标签以及预测的置信度；
匹配模块，用于根据所述预测的置信度确定所述待测的目标对象的承揽对象。
一种电子设备，包括：
一个或多个处理器；和其上存储有指令的一个或多个机器可读介质，当所述一个或多个处理器执行时，使得所述电子设备执行任一所述的承揽服务推荐方法。
一种机器可读介质，其上存储有指令，当由一个或多个处理器执行时，使得设备执行所述的承揽服务推荐方法。
附图说明
图1显示为本发明实施例的承揽服务推荐方法的示意图。
图2显示为本发明实施例的承揽服务推荐系统的示意图。
具体实施方式
以下通过特定的具体实例说明本发明的实施方式，本领域技术人员可由本说明书所揭露的内容轻易地了解本发明的其他优点与功效。本发明还可以通过另外不同的具体实施方式加以实施或应用，本说明书中的各项细节也可以基于不同观点与应用，在没有背离本发明的精神下进行各种修饰或改变。
需要说明的是，本实施例中所提供的图示仅以示意方式说明本发明的基本构想，遂图式中仅显示与本发明中有关的组件而非按照实际实施时的组件数目、形状及尺寸绘制，其实际实施时各组件的型态、数量及比例可为一种随意的改变，且其组件布局型态也可能更为复杂。本说明书所附图式所绘示的结构、比例、大小等，均仅用以配合说明书所揭示的内容，以供熟悉此技术的人士了解与阅读，并非用以限定本发明可实施的限定条件，故不具技术上的实质意义，任何结构的修饰、比例关系的改变或大小的调整，在不影响本发明所能产生的功效及所能达成的目的下，均应仍落在本发明所揭示的技术内容得能涵盖的范围内。同时，本说明书中所引用的如“上”、“下”、“左”、“右”、“中间”及“一”等的用语，亦仅为便于叙述的明了，而非用以限定本发明可实施的范围，其相对关系的改变或调整，在无实质变更技术内容下，当亦视为本发明可实施的范畴。
请参阅图1，本发明的其中一个实施例提供一种承揽服务推荐方法，包括：
S1：获取目标对象的目标文本和承揽对象的承揽文本，一般地，所述目标文本承载的信息与所述目标对象的需求相关联，同理，所述承揽文本的信息与所述承揽对象提供的承揽服务相关联，在承揽服务中，当目标对象的需求与承揽对象的承揽服务相匹配时，不仅能够满足目标对象的需求，而且能够便于承揽对象提供承揽服务，在承揽服务平台中，可以通过目标对象填写相关需求信息或者通过承揽对象填写承揽服务信息的方式获取相应的文本；
S2：将所述目标文本输入到第一神经网络中进行第一分类处理，获取处理后的目标标签，并将所述承揽文本输入到第二神经网络中进行第二分类处理，获得处理后的承揽标签，所述目标标签或者所述承揽标签为向量，将处理后的所述目标标签和处理后的所述承揽标签进行融合处理，获取融合标签，所述融合标签为特征矩阵，并将所述融合标签输入到第三神经网络中进行第三分类处理，获取与处理后的所述目标标签相匹配的承揽标签以及置信度，并通过训练所述第三神经网络获取分类模型，例如，通过迭代训练，检测目标标签与承揽标签匹配的准确率(Precision)和召回率(Recall)，又例如，检测置信度的F值(F-Measure)，获取优选的训练模型为分类模型；
S3：将待测的目标对象的目标标签和待测的承揽对象的承揽标签进行融合，得到待测的融合标签并输入到所述分类模型中，获取预测的承揽标签以及预测的置信度，获取目标标签和承揽标签的步骤包括：通过神经网络对文本进行分类处理获取标签，例如，通过第一神经网络处理获取目标标签，又例如，通过第二神经网络处理获取承揽标签；
S4：根据所述预测的置信度确定所述待测的目标对象的承揽对象，因此，所述承揽对象能够为所述目标对象提供较好的承揽服务。在承揽服务的推荐平台中，获取目标对象和承揽 对象的相关信息，并通过机器学习的方式学习目标对象和承揽对象的分类逻辑，通过模型训练获取具有该分类逻辑的分类模型，待测的目标对象通过所述分类模型的处理分配给相匹配的承揽对象。
在一些实施过程中，所述第一分类处理的步骤包括：将所述目标文本输入到所述第一神经网络中获取处理结果，并将所述处理结果与标注的目标标签进行相似度对比，根据相似度获取处理后的目标标签；
所述第二分类处理的步骤包括：将所述承揽文本输入到所述第二神经网络中获取处理结果，并将处理结果与标注的承揽标签进行相似度对比，根据相似度获取处理后的承揽标签。经过神经网络的分类或者聚类处理，从文本信息中获取标签信息，在实施过程中，可以将目标标签或者承揽标签进行向量化，例如，可以通过目标标签获取目标向量，又例如，可以通过承揽标签获取承揽向量，便于将目标向量和承揽向量进行融合。
进一步的，所述第三神经网络包括输入层、隐藏层以及输出层，通过损失函数对所述第三神经网络进行训练，所述损失函数包括：第一损失函数、第二损失函数以及第三损失函数，所述第一损失函数包括所述处理后的目标标签与所述标注的目标标签之间的损失，所述第二损失函数包括所述处理后的承揽标签与所述标注的承揽标签之间的损失，所述第三损失函数包括所述处理后的目标标签与所述预测的承揽标签之间的损失。
在一些实施过程中，损失函数的数学表达为：
L＝L1+L2+L3
Figure PCTCN2022114377-appb-000005
Figure PCTCN2022114377-appb-000006
Figure PCTCN2022114377-appb-000007
其中，L为损失函数，L1为第一损失函数，L2为第二损失函数，L3为第三损失函数，N为所述处理后的目标标签的数量，M为所述处理后的承揽标签的数量，
当所述处理后的目标标签的集合中第i个标签与标注的目标标签的集合中的第e个标签匹配时，x(ie)＝1，否则，x(ie)＝0；
当所述处理后的承揽标签的集合中第j个标签与标注的承揽标签的集合中的第f个标签匹配时，y(jf)＝1，否则，y(jf)＝0；
pie为所述处理后的目标标签的集合中第i个标签与标注的目标标签的集合中的第e个标签匹配的概率，pjf为所述处理后的承揽标签的集合中第j个标签与标注的承揽标签的集合中的第f个标签匹配的概率；
当处理后的目标标签与所述预测的承揽标签匹配时，zk＝1，否则，zk＝0；
pk为处理后的目标标签与所述预测的承揽标签匹配的概率。该损失函数不仅考虑所述处理后的目标标签与所述标注的目标标签之间的损失、所述处理后的承揽标签与所述标注的承揽标签之间的损失，还考虑了所述处理后的目标标签与所述预测的承揽标签之间的损失。通过该损失函数对第三神经网络以及数据进行训练，获取理想的分类模型，提高了分类模型的分类精度，并通过分类模型获取分类逻辑。
在一些实施过程中，通过所述预测的置信度确定期望参数，并通过所述期望参数确定所述待测的目标对象的承揽对象，所述期望参数的数学表达为：
Figure PCTCN2022114377-appb-000008
其中，Q为所述期望参数，C为所述待处理的目标标签的数量，a为第a个待处理的目标标签，D为所述预测的承揽标签的数量，b为第b个预测的承揽标签，Pa(Kb)为第a个待处理的目标标签所对应的第b个预测的承揽标签的置信度。例如，待测的目标对象具有C个目标标签，待测的承揽对象有多个，承揽对象具有D个承揽标签，计算D个承揽标签的置信度的加权值，进而衡量待测的承揽对象与待测的目标对象之间的匹配关系，通过期望参数的大小作为衡量匹配关系的指标。
进一步的，还可以用期望值来衡量上述匹配关系，例如，通过所述期望参数确定期望值，并通过各个所述承揽对象的所述期望值的排名，确定所述待测的目标对象的承揽对象，所述期望值的数学表达为：
W＝H Q
其中，W为所述期望值，H为大于1的常数，Q为所述期望参数。
为了便于对承揽标签进行分类，可将所述输出层的激活函数设置为sigmod函数。
请参阅图2，本发明提供一种承揽服务推荐系统，包括：
获取模块，用于获取目标对象的目标文本和承揽对象的承揽文本；
模型模块，用于将所述目标文本输入到第一神经网络中进行第一分类处理，获取处理后 的目标标签，并将所述承揽文本输入到第二神经网络中进行第二分类处理，获得处理后的承揽标签，将处理后的所述目标标签和处理后的所述承揽标签进行融合处理，获取融合标签，并将所述融合标签输入到第三神经网络中进行第三分类处理，获取与处理后的所述目标标签相匹配的承揽标签以及置信度，并通过训练第三神经网络获取分类模型；
处理模块，用于将待测的目标对象的目标标签和待测的承揽对象的承揽标签进行融合，得到待测的融合标签并输入到所述分类模型中，获取预测的承揽标签以及预测的置信度；
匹配模块，用于根据所述预测的置信度确定所述待测的目标对象的承揽对象。
进一步地，所述第一分类处理的步骤包括：将所述目标文本输入到所述第一神经网络中获取处理结果，并将所述处理结果与标注的目标标签进行相似度对比，根据相似度获取处理后的目标标签；
所述第二分类处理的步骤包括：将所述承揽文本输入到所述第二神经网络中获取处理结果，并将处理结果与标注的承揽标签进行相似度对比，根据相似度获取处理后的承揽标签。
进一步地，所述第三神经网络包括输入层、隐藏层以及输出层，通过损失函数对所述第三神经网络进行训练，所述损失函数包括第一损失函数、第二损失函数以及第三损失函数，所述第一损失函数包括所述处理后的目标标签与所述标注的目标标签之间的损失，所述第二损失函数包括所述处理后的承揽标签与所述标注的承揽标签之间的损失，所述第三损失函数包括所述处理后的目标标签与所述预测的承揽标签之间的损失。
进一步地，所述损失函数的数学表达为：
L＝L1+L2+L3
Figure PCTCN2022114377-appb-000009
Figure PCTCN2022114377-appb-000010
Figure PCTCN2022114377-appb-000011
其中，L为损失函数，L1为第一损失函数，L2为第二损失函数，L3为第三损失函数，N为所述处理后的目标标签的数量，M为所述处理后的承揽标签的数量，
当所述处理后的目标标签的集合中第i个标签与标注的目标标签的集合中的第e个标签 匹配时，x(ie)＝1，否则，x(ie)＝0；
当所述处理后的承揽标签的集合中第j个标签与标注的承揽标签的集合中的第f个标签匹配时，y(jf)＝1，否则，y(jf)＝0；
pie为所述处理后的目标标签的集合中第i个标签与标注的目标标签的集合中的第e个标签匹配的概率，pjf为所述处理后的承揽标签的集合中第j个标签与标注的承揽标签的集合中的第f个标签匹配的概率；
当处理后的目标标签与所述预测的承揽标签匹配时，zk＝1，否则，zk＝0；
pk为处理后的目标标签与所述预测的承揽标签匹配的概率。
进一步地，通过所述预测的置信度确定期望参数，并通过所述期望参数确定所述待测的目标对象的承揽对象，所述期望参数的数学表达为：
Figure PCTCN2022114377-appb-000012
其中，Q为所述期望参数，C为所述待测的目标标签的数量，a为第a个待测的目标标签，D为所述预测的承揽标签的数量，b为第b个预测的承揽标签，Pa(Kb)为第a个待测的目标标签所对应的第b个预测的承揽标签的置信度。
进一步地，通过所述期望参数确定期望值，并通过各个所述承揽对象的所述期望值的排名，确定所述待测的目标对象的承揽对象，所述期望值的数学表达为：
W＝H Q
其中，W为所述期望值，H为大于1的常数，Q为所述期望参数。
进一步地，所述输出层的激活函数为sigmod函数。
本发明实施例提供一种电子设备，包括：一个或多个处理器；和其上存储有指令的一个或多个机器可读介质，当所述一个或多个处理器执行时，使得所述电子设备执行一个或多个所述的方法。本发明可用于众多通用或专用的计算系统环境或配置中。例如：个人计算机、服务器计算机、手持设备或便携式设备、平板型设备、多处理器系统、基于微处理器的系统、置顶盒、可编程的消费电子设备、网络PC、小型计算机、大型计算机、包括以上任何系统或设备的分布式计算环境等等。
本发明实施例还提供一个或多个机器可读介质，其上存储有指令，当由一个或多个处理器执行时，使得设备执行中一个或多个所述的方法。本发明可以在由计算机执行的计算机可执行指令的一般上下文中描述，例如程序模块。一般地，程序模块包括执行特定任务或实现特定抽象数据类型的例程、程序、对象、组件、数据结构等等。也可以在分布式计算环境中 实践本发明，在这些分布式计算环境中，由通过通信网络而被连接的远程处理设备来执行任务。在分布式计算环境中，程序模块可以位于包括存储设备在内的本地和远程计算机存储介质中。
如上所述，本发明的承揽服务推荐方法及系统，具有以下有益效果：
在承揽服务的推荐平台中，获取目标对象和承揽对象的相关信息，并通过机器学习的方式学习目标对象和承揽对象的分类逻辑，通过模型训练获取具有该分类逻辑的分类模型，待测的目标对象通过所述分类模型的处理分配给相匹配的承揽对象。
上述实施例仅例示性说明本发明的原理及其功效，而非用于限制本发明。任何熟悉此技术的人士皆可在不违背本发明的精神及范畴下，对上述实施例进行修饰或改变。因此，举凡所属技术领域中具有通常知识者在未脱离本发明所揭示的精神与技术思想下所完成的一切等效修饰或改变，仍应由本发明的权利要求所涵盖。

Claims (10)

  1. 一种承揽服务推荐方法，其特征在于，包括：
    获取目标对象的目标文本和承揽对象的承揽文本；
    将所述目标文本输入到第一神经网络中进行第一分类处理，获取处理后的目标标签，并将所述承揽文本输入到第二神经网络中进行第二分类处理，获得处理后的承揽标签，将处理后的所述目标标签和处理后的所述承揽标签进行融合处理，获取融合标签，并将所述融合标签输入到第三神经网络中进行第三分类处理，获取与处理后的所述目标标签相匹配的承揽标签以及置信度，并通过训练所述第三神经网络获取分类模型；
    将待测的目标对象的目标标签和待测的承揽对象的承揽标签进行融合，得到待测的融合标签并输入到所述分类模型中，获取预测的承揽标签以及预测的置信度；
    根据所述预测的置信度确定所述待测的目标对象的承揽对象。
  2. 根据权利要求1所述的承揽服务推荐方法，其特征在于，所述第一分类处理的步骤包括：将所述目标文本输入到所述第一神经网络中获取处理结果，并将所述处理结果与标注的目标标签进行相似度对比，根据相似度获取处理后的目标标签；
    所述第二分类处理的步骤包括：将所述承揽文本输入到所述第二神经网络中获取处理结果，并将处理结果与标注的承揽标签进行相似度对比，根据相似度获取处理后的承揽标签。
  3. 根据权利要求2所述的承揽服务推荐方法，其特征在于，所述第三神经网络包括输入层、隐藏层以及输出层，通过损失函数对所述第三神经网络进行训练，所述损失函数包括第一损失函数、第二损失函数以及第三损失函数，所述第一损失函数包括所述处理后的目标标签与所述标注的目标标签之间的损失，所述第二损失函数包括所述处理后的承揽标签与所述标注的承揽标签之间的损失，所述第三损失函数包括所述处理后的目标标签与所述预测的承揽标签之间的损失。
  4. 根据权利要求3所述的承揽服务推荐方法，其特征在于，所述损失函数的数学表达为：
    L＝L1+L2+L3
    Figure PCTCN2022114377-appb-100001
    Figure PCTCN2022114377-appb-100002
    Figure PCTCN2022114377-appb-100003
    其中，L为损失函数，L1为第一损失函数，L2为第二损失函数，L3为第三损失函数，N为所述处理后的目标标签的数量，M为所述处理后的承揽标签的数量，
    当所述处理后的目标标签的集合中第i个标签与标注的目标标签的集合中的第e个标签匹配时，x(ie)＝1，否则，x(ie)＝0；
    当所述处理后的承揽标签的集合中第j个标签与标注的承揽标签的集合中的第f个标签匹配时，y(jf)＝1，否则，y(jf)＝0；
    pie为所述处理后的目标标签的集合中第i个标签与标注的目标标签的集合中的第e个标签匹配的概率，pjf为所述处理后的承揽标签的集合中第j个标签与标注的承揽标签的集合中的第f个标签匹配的概率；
    当处理后的目标标签与所述预测的承揽标签匹配时，zk＝1，否则，zk＝0；
    pk为处理后的目标标签与所述预测的承揽标签匹配的概率。
  5. 根据权利要求1所述的承揽服务推荐方法，其特征在于，通过所述预测的置信度确定期望参数，并通过所述期望参数确定所述待测的目标对象的承揽对象，所述期望参数的数学表达为：
    Figure PCTCN2022114377-appb-100004
    其中，Q为所述期望参数，C为所述待测的目标标签的数量，a为第a个待测的目标标签，D为所述预测的承揽标签的数量，b为第b个预测的承揽标签，Pa(Kb)为第a个待测的目标标签所对应的第b个预测的承揽标签的置信度。
  6. 根据权利要求5所述的承揽服务推荐方法，其特征在于，通过所述期望参数确定期望值，并通过各个所述承揽对象的所述期望值的排名，确定所述待测的目标对象的承揽对象，所述期望值的数学表达为：
    W＝H Q
    其中，W为所述期望值，H为大于1的常数，Q为所述期望参数。
  7. 根据权利要求3所述的承揽服务推荐方法，其特征在于，所述输出层的激活函数为sigmod函数。
  8. 一种承揽服务推荐系统，其特征在于，包括：
    获取模块，用于获取目标对象的目标文本和承揽对象的承揽文本；
    模型模块，用于将所述目标文本输入到第一神经网络中进行第一分类处理，获取处理后的目标标签，并将所述承揽文本输入到第二神经网络中进行第二分类处理，获得处理后的承揽标签，将处理后的所述目标标签和处理后的所述承揽标签进行融合处理，获取融合标签，并将所述融合标签输入到第三神经网络中进行第三分类处理，获取与处理后的所述目标标签相匹配的承揽标签以及置信度，并通过训练所述第三神经网络获取分类模型；
    处理模块，用于将待测的目标对象的目标标签和待测的承揽对象的承揽标签进行融合，得到待测的融合标签并输入到所述分类模型中，获取预测的承揽标签以及预测的置信度；
    匹配模块，用于根据所述预测的置信度确定所述待测的目标对象的承揽对象。
  9. 一种电子设备，其特征在于，包括：
    一个或多个处理器；和
    其上存储有指令的一个或多个机器可读介质，当所述一个或多个处理器执行时，使得所述电子设备执行如权利要求1-7中任一所述的承揽服务推荐方法。
  10. 一种机器可读介质，其特征在于，其上存储有指令，当由一个或多个处理器执行时，使得设备执行如权利要求1-7中任一所述的承揽服务推荐方法。
PCT/CN2022/114377 2021-09-10 2022-08-24 一种承揽服务推荐方法及系统 WO2023035941A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111059416.7A CN113505230B (zh) 2021-09-10 2021-09-10 一种承揽服务推荐方法及系统
CN202111059416.7 2021-09-10

Publications (1)

Publication Number Publication Date
WO2023035941A1 true WO2023035941A1 (zh) 2023-03-16

Family

ID=78016602

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/114377 WO2023035941A1 (zh) 2021-09-10 2022-08-24 一种承揽服务推荐方法及系统

Country Status (2)

Country Link
CN (1) CN113505230B (zh)
WO (1) WO2023035941A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113505230B (zh) * 2021-09-10 2021-12-21 明品云(北京)数据科技有限公司 一种承揽服务推荐方法及系统

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018161710A1 (zh) * 2017-03-07 2018-09-13 广州优视网络科技有限公司 一种基于输入搜索词来推荐搜索词的方法、装置和存储介质
CN109522406A (zh) * 2018-10-11 2019-03-26 平安科技（深圳）有限公司 文本语义匹配方法、装置、计算机设备和存储介质
CN110969015A (zh) * 2019-11-28 2020-04-07 国网上海市电力公司 一种基于运维脚本的标签自动化识别方法和设备
CN111382352A (zh) * 2020-03-02 2020-07-07 腾讯科技（深圳）有限公司 数据推荐方法、装置、计算机设备以及存储介质
US20210035183A1 (en) * 2019-07-31 2021-02-04 Synchronoss Technologies, Inc. Method and system for a recommendation engine utilizing progressive labeling and user content enrichment
CN113505230A (zh) * 2021-09-10 2021-10-15 明品云(北京)数据科技有限公司 一种承揽服务推荐方法及系统

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113508378A (zh) * 2019-10-31 2021-10-15 华为技术有限公司 推荐模型的训练方法、推荐方法、装置及计算机可读介质
CN113076405A (zh) * 2021-06-04 2021-07-06 明品云(北京)数据科技有限公司 基于用户画像的解决方案匹配方法及系统
CN113360777B (zh) * 2021-08-06 2021-12-07 北京达佳互联信息技术有限公司 内容推荐模型训练方法、内容推荐方法及相关设备

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018161710A1 (zh) * 2017-03-07 2018-09-13 广州优视网络科技有限公司 一种基于输入搜索词来推荐搜索词的方法、装置和存储介质
CN109522406A (zh) * 2018-10-11 2019-03-26 平安科技（深圳）有限公司 文本语义匹配方法、装置、计算机设备和存储介质
US20210035183A1 (en) * 2019-07-31 2021-02-04 Synchronoss Technologies, Inc. Method and system for a recommendation engine utilizing progressive labeling and user content enrichment
CN110969015A (zh) * 2019-11-28 2020-04-07 国网上海市电力公司 一种基于运维脚本的标签自动化识别方法和设备
CN111382352A (zh) * 2020-03-02 2020-07-07 腾讯科技（深圳）有限公司 数据推荐方法、装置、计算机设备以及存储介质
CN113505230A (zh) * 2021-09-10 2021-10-15 明品云(北京)数据科技有限公司 一种承揽服务推荐方法及系统

Also Published As

Publication number Publication date
CN113505230B (zh) 2021-12-21
CN113505230A (zh) 2021-10-15

Similar Documents

Publication Publication Date Title
US11250487B2 (en) Computer vision and image characteristic search
US11714831B2 (en) Data processing and classification
CN107220365B (zh) 基于协同过滤与关联规则并行处理的精准推荐系统及方法
US9384278B2 (en) Methods and systems for assessing excessive accessory listings in search results
US11017016B2 (en) Clustering product media files
CN110580489B (zh) 一种数据对象的分类系统、方法以及设备
US11367117B1 (en) Artificial intelligence system for generating network-accessible recommendations with explanatory metadata
WO2023035941A1 (zh) 一种承揽服务推荐方法及系统
CN112667825A (zh) 基于知识图谱的智能推荐方法、装置、设备及存储介质
US20190066054A1 (en) Accuracy of member profile retrieval using a universal concept graph
CN111814056A (zh) 基于信息处理的供应商推荐方法及相关设备
WO2022262752A1 (zh) 基于数据交互的信息推荐方法、装置、设备及存储介质
CN110991538B (zh) 样本分类方法、装置、存储介质和计算机设备
CN113343101A (zh) 一种对象排序方法及系统
CN115374354A (zh) 基于机器学习的科技服务推荐方法、装置、设备及介质
WO2023035940A1 (zh) 一种目标对象推荐方法及系统
CN112948705A (zh) 一种基于政策大数据的智能匹配方法、装置及介质
US20050207653A1 (en) Method for analysis of line objects
WO2022262775A1 (zh) 基于数据交互的信息处理方法、装置、设备及存储介质
CN115982597A (zh) 语义相似度模型训练方法及装置、语义匹配方法及装置
CN111488479A (zh) 超图构建方法、装置以及计算机系统和介质
CN109886299A (zh) 一种用户画像方法、装置、可读存储介质及终端设备
CN110827101A (zh) 一种店铺推荐的方法和装置
CN110262906B (zh) 接口标签推荐方法、装置、存储介质和电子设备
CN112966013B (zh) 知识展示方法、装置、设备及可读存储介质

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: 22866411

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE