CN110543558B - Problem matching method, device, equipment and medium - Google Patents
Problem matching method, device, equipment and mediumDownload PDFInfo
- Publication number
- CN110543558B CN110543558BCN201910843848.3ACN201910843848ACN110543558BCN 110543558 BCN110543558 BCN 110543558BCN 201910843848 ACN201910843848 ACN 201910843848ACN 110543558 BCN110543558 BCN 110543558B
- Authority
- CN
- China
- Prior art keywords
- question
- network layer
- sample
- classification
- similarity
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Databases & Information Systems (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
Description
Translated fromChinese技术领域technical field
本申请实施例涉及数据处理领域，尤其涉及智能搜索技术。具体地，本实施例涉及了一种问题匹配方法、装置、设备和介质。The embodiments of the present application relate to the field of data processing, and in particular to intelligent search technology. Specifically, this embodiment relates to a question matching method, device, device and medium.
背景技术Background technique
现有应用于问题匹配的模型往往是基于分类任务或者基于排序任务进行训练的。在分类任务中，给定两条问题，需要训练模型来判断这两条问题是否相似；在排序任务中，给定一个中心问题，一个正样本和一个负样本，需要模型给正样本的分数高于负样本的分数。Existing models applied to question matching are often trained based on classification tasks or ranking tasks. In the classification task, given two questions, it is necessary to train the model to judge whether the two questions are similar; in the sorting task, given a central question, a positive sample and a negative sample, the model needs to give the positive sample a high score on the score of negative samples.
在排序任务中，现有应用于问题匹配的模型存在如下缺陷：In the ranking task, existing models applied to question matching suffer from the following shortcomings:
使用分类任务训练的分类模型，忽略了相似程度的信息，因此使用分类模型进行排序的效果不理想。The classification model trained by the classification task ignores the information of the degree of similarity, so the effect of using the classification model for sorting is not ideal.
使用排序任务训练的排序模型，虽然考虑到了相似程度的不同，但在构造数据集的过程中往往使用随机采样方法，该方法涵盖的负样本空间较小(尤其在数据较多、负样本空间较大的情况下)，从而使得训练的排序模型的鲁棒性不够，或者有一定的偏向性。Although the sorting model trained by the sorting task takes into account the difference in similarity, the random sampling method is often used in the process of constructing the data set, which covers a small negative sample space (especially when there is more data and the negative sample space is relatively large). In the case of large size), the robustness of the trained ranking model is not enough, or there is a certain bias.
发明内容Contents of the invention
本申请实施例提供一种问题匹配方法、装置、设备和介质，以提高问题排序模型的鲁棒性，进而提高问题匹配的准确率。Embodiments of the present application provide a question matching method, device, device, and medium, so as to improve the robustness of a question ranking model, thereby improving the accuracy of question matching.
第一方面，本申请实施例提供了一种问题匹配方法，该方法包括：In the first aspect, the embodiment of the present application provides a question matching method, which includes:
利用问题分类样本，对问题分类模型与问题排序模型复用的基础网络层，以及问题分类模型中的分类输出网络层进行训练；Use the problem classification samples to train the basic network layer reused by the problem classification model and the problem ranking model, as well as the classification output network layer in the problem classification model;
利用问题排序样本，对所述基础网络层以及问题排序模型中的排序输出网络层进行训练，以得到经训练的问题排序模型供进行问题匹配。Using the question ranking samples, the basic network layer and the ranking output network layer in the question ranking model are trained to obtain a trained question ranking model for question matching.
本申请实施例具有如下优点或有益效果：通过在利用问题排序样本，对所述基础网络层以及问题排序模型中的排序输出网络层进行训练的基础上，利用问题分类样本，对问题分类模型与问题排序模型复用的基础网络层进行训练。因为增加了问题分类样本对基础网络层的训练，所以可以增大问题排序模型的负样本空间，从而提高问题排序模型的鲁棒性，减少问题排序模型的偏向性。The embodiment of the present application has the following advantages or beneficial effects: on the basis of using the question ranking samples to train the basic network layer and the ranking output network layer in the question ranking model, using the question classification samples to train the question classification model and The question ranking model reuses the base network layer for training. Because the training of the basic network layer with question classification samples is added, the negative sample space of the question ranking model can be increased, thereby improving the robustness of the question ranking model and reducing the bias of the question ranking model.
可选地，在每一轮训练过程中，利用该轮问题分类样本和该轮问题排序样本分别进行训练。Optionally, during each round of training, the round of question classification samples and the round of question sorting samples are used for training respectively.
基于该技术特征，本申请实施例具有如下优点或有益效果：通过在每一轮训练过程中，利用该轮问题分类样本和该轮问题排序样本分别进行训练，从而实现利用问题分类样本和问题排序样本对问题排序模型的交叉训练，进而减少模型偏向性。Based on this technical feature, the embodiment of the present application has the following advantages or beneficial effects: by using the round of question classification samples and the round of question ranking samples for training in each round of training process, so as to realize the use of question classification samples and question ranking Cross-training of sample-to-question ranking models to reduce model bias.
可选地，在利用问题分类样本训练过程中，采用分类损失函数；Optionally, a classification loss function is used during the training process of using the problem classification samples;
在利用问题排序样本训练过程中，采用排序损失函数。In the training process of ranking samples using questions, a ranking loss function is used.
基于该技术特征，本申请实施例具有如下优点或有益效果：通过利用问题分类样本训练过程中，采用分类损失函数；在利用问题排序样本训练过程中，采用排序损失函数，从而提高问题排序模型的训练准确性，进而提高问题排序模型的鲁棒性。Based on this technical feature, the embodiment of the present application has the following advantages or beneficial effects: by using the classification loss function in the training process of using problem classification samples; in the training process of using problem ranking samples, the ranking loss function is used, thereby improving the performance of the problem ranking model. Training accuracy, which in turn improves the robustness of the question ranking model.
可选地，所述问题分类样本包括正样本问句对和负样本问句对；Optionally, the question classification samples include positive sample question pairs and negative sample question pairs;
问题排序样本包括样本三元组，其中该三元组包括中心问句、正样问句和负样问句，所述正样问句与所述中心问句的相似度大于第一相似阈值，所述负样问句与所述中心问句的相似度小于第二相似阈值，所述第一相似阈值大于或等于所述第二相似阈值。The question sorting sample includes a sample triplet, wherein the triplet includes a central question, a positive question and a negative question, and the similarity between the positive question and the central question is greater than the first similarity threshold, The similarity between the negative sample question and the central question is smaller than a second similarity threshold, and the first similarity threshold is greater than or equal to the second similarity threshold.
基于该技术特征，本申请实施例具有如下优点或有益效果：利用正样本问句对、负样本问句对和样本三元组，对问题排序模型的基础网络层进行分类和排序的联合训练。Based on this technical feature, the embodiments of the present application have the following advantages or beneficial effects: use positive sample question pairs, negative sample question pairs and sample triples to perform joint training of classification and sorting on the basic network layer of the question ranking model.
可选地，上述方法还包括：Optionally, the above method also includes:
得到经训练的问题排序模型之后，将用户输入的待检索问句和已有问答对中已有问句作为经训练的问题排序模型的输入，以得到所述待检索问句与所述已有问句之间的语义相似度；After the trained question ranking model is obtained, the user-input question sentence to be retrieved and the existing question sentence in the existing question-answer pair are used as the input of the trained question ranking model to obtain the question sentence to be retrieved and the existing question sentence. Semantic similarity between questions;
根据所述待检索问句与所述已有问句之间的语义相似度，确定与所述待检索问句匹配的目标问句以及目标答案。According to the semantic similarity between the question sentence to be retrieved and the existing question sentence, a target question sentence and a target answer matching the question sentence to be retrieved are determined.
基于该技术特征，本申请实施例具有如下优点或有益效果：通过利用经分类和排序联合训练的问题排序模型，从而实现对待检索问句与已有问句之间语义相似度的准确确定，进而提高目标问句和目标答案的确定准确率。Based on this technical feature, the embodiment of the present application has the following advantages or beneficial effects: by using the question ranking model trained jointly by classification and ranking, the accurate determination of the semantic similarity between the question to be retrieved and the existing question is realized, and then Improve the determination accuracy of target questions and target answers.
可选地，所述问题分类样本根据通用数据构建；Optionally, the problem classification sample is constructed according to general data;
所述问题排序样本根据目标领域数据构建。The question ranking sample is constructed according to the target field data.
基于该技术特征，本申请实施例具有如下优点或有益效果：通过根据通用数据构建问题分类样本，从而使得基于该问题分类样本训练得到的问题排序模型能够学习到通用数据的语义关系。通过根据目标领域数据构建问题排序样本，从而使得基于该问题排序样本训练得到的问题排序模型学习到针对目标领域数据的语义关系，进而提高问题排序模型的排序准确率。Based on this technical feature, the embodiments of the present application have the following advantages or beneficial effects: by constructing question classification samples based on general data, the question ranking model trained based on the question classification samples can learn the semantic relationship of general data. By constructing a question ranking sample based on the target field data, the question ranking model trained based on the question ranking sample can learn the semantic relationship for the target field data, thereby improving the ranking accuracy of the question ranking model.
第二方面，本申请实施例提供了一种问题匹配装置，该装置包括：In the second aspect, the embodiment of the present application provides a question matching device, which includes:
分类训练模块，用于利用问题分类样本，对问题分类模型与问题排序模型复用的基础网络层，以及问题分类模型中的分类输出网络层进行训练；The classification training module is used to use the problem classification samples to train the basic network layer reused by the problem classification model and the problem ranking model, and the classification output network layer in the problem classification model;
排序训练模块，用于利用问题排序样本，对所述基础网络层以及问题排序模型中的排序输出网络层进行训练，以得到经训练的问题排序模型供进行问题匹配。The ranking training module is used to use the question ranking samples to train the basic network layer and the ranking output network layer in the question ranking model, so as to obtain a trained question ranking model for question matching.
第三方面，本申请实施例提供了一种电子设备，该设备包括：In a third aspect, an embodiment of the present application provides an electronic device, which includes:
至少一个处理器；以及at least one processor; and
与所述至少一个处理器通信连接的存储器；其中，a memory communicatively coupled to the at least one processor; wherein,
所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行本申请实施例中任一项所述的问题匹配方法。The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor, so that the at least one processor can execute the method described in any one of the embodiments of the present application. Question matching method.
第四方面，本申请实施例提供了一种存储有计算机指令的非瞬时计算机可读存储介质，所述计算机指令用于使所述计算机执行本申请实施例中任一项所述的问题匹配方法。In a fourth aspect, the embodiments of the present application provide a non-transitory computer-readable storage medium storing computer instructions, the computer instructions are used to make the computer execute the question matching method described in any one of the embodiments of the present application .
上述可选方式所具有的其他效果将在下文中结合具体实施例加以说明。Other effects of the above optional manner will be described below in conjunction with specific embodiments.
附图说明Description of drawings
附图用于更好地理解本方案，不构成对本申请的限定。其中：The accompanying drawings are used to better understand the solution, and do not constitute a limitation to the application. in:
图1是本申请第一实施例提供的一种问题匹配方法的流程图；FIG. 1 is a flow chart of a question matching method provided in the first embodiment of the present application;
图2为本申请第二实施例提供的一种基础网络层的结构示意图；FIG. 2 is a schematic structural diagram of a basic network layer provided in a second embodiment of the present application;
图3为本申请第二实施例提供的分类任务训练的模型结构示意图；Fig. 3 is a schematic diagram of the model structure of the classification task training provided by the second embodiment of the present application;
图4为本申请第二实施例提供的排序任务训练的模型结构示意图；FIG. 4 is a schematic diagram of the model structure of the sorting task training provided by the second embodiment of the present application;
图5是本申请第三实施例提供的一种问题匹配装置的结构示意图；Fig. 5 is a schematic structural diagram of a question matching device provided in the third embodiment of the present application;
图6是本申请第四实施例提供的问题匹配方法的电子设备的框图。Fig. 6 is a block diagram of an electronic device for a question matching method provided in a fourth embodiment of the present application.
具体实施方式detailed description
以下结合附图对本申请的示范性实施例做出说明，其中包括本申请实施例的各种细节以助于理解，应当将它们认为仅仅是示范性的。因此，本领域普通技术人员应当认识到，可以对这里描述的实施例做出各种改变和修改，而不会背离本申请的范围和精神。同样，为了清楚和简明，以下的描述中省略了对公知功能和结构的描述。Exemplary embodiments of the present application are described below in conjunction with the accompanying drawings, which include various details of the embodiments of the present application to facilitate understanding, and they should be regarded as exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the application. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.
第一实施例first embodiment
图1是本申请第一实施例提供的一种问题匹配方法的流程图。本实施例可适用于确定两问题之间相似度的情况。典型地，本实施例适用于检索式问答系统中确定检索问题与已有问答对中已有问题之间相似度的情况。该方法可以由软件和/或硬件的方式实现。参见图1，本实施例提供的问题匹配方法包括：Fig. 1 is a flow chart of a question matching method provided in the first embodiment of the present application. This embodiment is applicable to the situation of determining the similarity between two questions. Typically, this embodiment is applicable to the situation of determining the similarity between the retrieval question and the existing question in the existing question-answer pair in the retrieval question answering system. The method can be implemented by means of software and/or hardware. Referring to Fig. 1, the problem matching method provided in this embodiment includes:
S110、利用问题分类样本，对问题分类模型与问题排序模型复用的基础网络层，以及问题分类模型中的分类输出网络层进行训练。S110. Using the question classification samples, train the basic network layer reused by the question classification model and the question ranking model, and the classification output network layer in the question classification model.
其中，问题分类样本是指对模型进行问题分类任务训练的样本。Wherein, the problem classification sample refers to a sample for training the model on the problem classification task.
具体地，问题分类样本包括正样本问句对和负样本问句对。Specifically, the question classification samples include positive sample question pairs and negative sample question pairs.
示例性地，正样本问句对是指相似度大于第三相似阈值的两个问句。负样本问句是指相似度小于第四相似阈值的两个问句。其中第四相似阈值小于或等于第三相似阈值。Exemplarily, a positive sample question pair refers to two question sentences whose similarity is greater than a third similarity threshold. Negative sample questions refer to two questions whose similarity is less than the fourth similarity threshold. Wherein the fourth similarity threshold is less than or equal to the third similarity threshold.
问题分类模型是对问题进行分类的模型。A question classification model is a model that classifies questions.
问题排序模型是对问题进行相似度确定的模型，基于该模型可以实现根据问题对之间的相似度，对问题的排序。The question sorting model is a model for determining the similarity of questions. Based on this model, the question can be sorted according to the similarity between question pairs.
问题分类模型与问题排序模型复用的基础网络层是指，除输出网络层以外的其他网络层，也即问题分类模型与问题排序模型共用除输出网络层以外的其他网络层。The basic network layer reused by the problem classification model and the problem ranking model refers to other network layers except the output network layer, that is, the problem classification model and the problem ranking model share other network layers except the output network layer.
具体地，利用问题分类样本，对问题分类模型与问题排序模型复用的基础网络层，以及问题分类模型中的分类输出网络层进行训练，包括：Specifically, using the problem classification samples, the basic network layer reused by the problem classification model and the problem ranking model, and the classification output network layer in the problem classification model are trained, including:
将问题分类样本输入基础网络层中的输入网络层，经基础网络层中输入网络层外的其他网络层的处理，进入分类输出网络层；Input the problem classification sample into the input network layer in the basic network layer, and enter the classification output network layer after being processed by other network layers other than the input network layer in the basic network layer;
基于分类损失函数，根据分类输出网络层的输出结果和问题分类样本中的分类标签，调整基础网络层和分类输出网络层中的待训练参数。Based on the classification loss function, the parameters to be trained in the basic network layer and the classification output network layer are adjusted according to the output result of the classification output network layer and the classification label in the problem classification sample.
S120、利用问题排序样本，对所述基础网络层以及问题排序模型中的排序输出网络层进行训练，以得到经训练的问题排序模型供进行问题匹配。S120. Using the question ranking samples, train the basic network layer and the ranking output network layer in the question ranking model to obtain a trained question ranking model for question matching.
其中，所述问题排序样本是指对模型进行问题排序任务训练的样本。Wherein, the question ranking samples refer to samples for training the model on the question ranking task.
具体地，问题排序样本包括样本三元组，其中该三元组包括中心问句、正样问句和负样问句，所述正样问句与所述中心问句的相似度大于第一相似阈值，所述负样问句与所述中心问句的相似度小于第二相似阈值，所述第一相似阈值大于或等于所述第二相似阈值。第一相似阈值可以等于第三相似阈值，第二相似阈值可以等于第四相似阈值。Specifically, the question ranking sample includes a sample triplet, wherein the triplet includes a central question, a positive question, and a negative question, and the similarity between the positive question and the central question is greater than the first A similarity threshold, the similarity between the negative sample question and the central question is smaller than a second similarity threshold, and the first similarity threshold is greater than or equal to the second similarity threshold. The first similarity threshold may be equal to the third similarity threshold, and the second similarity threshold may be equal to the fourth similarity threshold.
利用问题排序样本，对所述基础网络层以及问题排序模型中的排序输出网络层进行训练，包括：Utilize the question sorting sample to train the basic network layer and the sorting output network layer in the question ranking model, including:
将中心问句和正样问句输入基础网络层中的输入网络层，经基础网络层中输入网络层外的其他网络层的处理，进入排序输出网络层，输出正样问句的相似度；Input the central question sentence and the correct question sentence into the input network layer in the basic network layer, and enter the sorting output network layer through the processing of other network layers outside the input network layer in the basic network layer, and output the similarity of the correct question sentence;
将中心问句和负样问句输入基础网络层中的输入网络层，经基础网络层中输入网络层外的其他网络层的处理，进入排序输出网络层，输出负样问句的相似度；Input the central question sentence and the negative sample question sentence into the input network layer in the basic network layer, and enter the sorting output network layer through the processing of other network layers outside the input network layer in the basic network layer, and output the similarity of the negative sample question sentence;
基于排序损失函数，根据输出正样问句的相似度和负样问句的相似度的大小关系，调整基础网络层和分类输出网络层中的待训练参数。Based on the ranking loss function, the parameters to be trained in the basic network layer and the classification output network layer are adjusted according to the relationship between the similarity of the output positive questions and the similarity of the negative questions.
本申请实施例的技术方案，通过在利用问题排序样本，对所述基础网络层以及问题排序模型中的排序输出网络层进行训练的基础上，利用问题分类样本，对问题分类模型与问题排序模型复用的基础网络层进行训练。因为增加了问题分类样本对基础网络层的训练，所以可以增大问题排序模型的负样本空间，从而提高问题排序模型的鲁棒性，减少问题排序模型的偏向性。In the technical solution of the embodiment of the present application, on the basis of using the problem sorting samples to train the basic network layer and the sorting output network layer in the problem sorting model, using the problem classification samples to train the problem classification model and the problem ranking model Reuse the base network layer for training. Because the training of the basic network layer with question classification samples is added, the negative sample space of the question ranking model can be increased, thereby improving the robustness of the question ranking model and reducing the bias of the question ranking model.
可选地，在每一轮训练过程中，利用该轮问题分类样本和该轮问题排序样本分别进行训练。Optionally, during each round of training, the round of question classification samples and the round of question sorting samples are used for training respectively.
也即，将训练过程分为多轮，每轮中利用一个问题分类样本集和一个问题排序样本集分别进行训练。That is, the training process is divided into multiple rounds, and in each round, a sample set of question classification and a sample set of question sorting are used for training respectively.
本实施例对问题分类样本集和问题排序样本集的训练顺序不作限定。可选地，可以先利用问题分类样本集训练，也可以先利用问题排序样本集进行训练。In this embodiment, there is no limitation on the training sequence of the question classification sample set and question sorting sample set. Optionally, the problem classification sample set may be used for training first, or the problem sorting sample set may be used for training first.
基于该技术特征，本申请实施例通过在每一轮训练过程中，利用该轮问题分类样本和该轮问题排序样本分别进行训练，从而实现利用问题分类样本和问题排序样本对问题排序模型的交叉训练，进而减少模型偏向性。Based on this technical feature, the embodiments of the present application use the round of question classification samples and the round of question ranking samples to perform training respectively during each round of training, so as to realize the crossover of the question ranking model by using the question classification samples and question ranking samples. training, thereby reducing model bias.
为实现基于待检索问句与已有问句之间语义相似度，对已有问句的排序，以确定目标问句，在得到经训练的问题排序模型之后，本实施例所述的方法还包括：In order to realize the sorting of the existing questions based on the semantic similarity between the questions to be retrieved and the existing questions, to determine the target questions, after obtaining the trained question sorting model, the method described in this embodiment also include:
将用户输入的待检索问句和已有问答对中已有问句作为经训练的问题排序模型的输入，以得到所述待检索问句与所述已有问句之间的语义相似度；Using the question sentence to be retrieved input by the user and the existing question sentence in the existing question and answer pair as the input of the trained question ranking model, to obtain the semantic similarity between the question sentence to be retrieved and the existing question sentence;
根据所述待检索问句与所述已有问句之间的语义相似度，确定与所述待检索问句匹配的目标问句以及目标答案。According to the semantic similarity between the question sentence to be retrieved and the existing question sentence, a target question sentence and a target answer matching the question sentence to be retrieved are determined.
其中，目标问句是与待检索问句匹配成功的问句，目标答案是目标问句关联的答案。Wherein, the target question sentence is a question sentence that is successfully matched with the question sentence to be retrieved, and the target answer is an answer associated with the target question sentence.
为使得问题排序模型能够学习到通用数据的语义关系和目标领域数据的语义关系，所述问题分类样本根据通用数据构建；In order to enable the problem ranking model to learn the semantic relationship of general data and the semantic relationship of target field data, the problem classification samples are constructed according to general data;
所述问题排序样本根据目标领域数据构建。The question ranking sample is constructed according to the target domain data.
第二实施例second embodiment
本实施例是在上述实施例的基础上提出的一种可选方案。This embodiment is an optional solution proposed on the basis of the foregoing embodiments.
图2为基础网络层的结构示意图。FIG. 2 is a schematic diagram of the structure of the basic network layer.
如图2所示，给定两个输入问句，确定问句中各词语的词向量，并且将输入问句投射为一个句向量；As shown in Figure 2, given two input questions, determine the word vector of each word in the question, and project the input question into a sentence vector;
根据两个问句的句向量确定组合向量，将组合向量输入全链接层，经全链接层处理，并输入分类输出网络层或排序输出网络层。The combination vector is determined according to the sentence vectors of the two questions, and the combination vector is input into the full link layer, processed by the full link layer, and input into the classification output network layer or the sorting output network layer.
为提高相似度确定的准确率，在组合向量中融入更多相似度描述元素。In order to improve the accuracy of similarity determination, more similarity description elements are incorporated into the combination vector.
典型地，根据两问句的句向量确定组合向量，包括：Typically, the combination vector is determined according to the sentence vectors of the two questions, including:
拼接两问句的句向量；Splicing the sentence vector of the two questions;
计算两问句的句向量的差，并对差值取模；Calculate the difference between the sentence vectors of the two questions, and take the modulo of the difference;
计算两问句的句向量的乘积；Calculate the product of the sentence vectors of the two questions;
将拼接结果、取模结果和乘积结果组合，生成组合向量。Combine the concatenation result, modulo result, and product result to generate a combined vector.
该基础网络层具有两个输出分支，分别为分类任务训练分支和排序任务训练分支，两分支共享权重。The basic network layer has two output branches, namely a classification task training branch and a sorting task training branch, and the two branches share weights.
具体训练任务描述如下：The specific training tasks are described as follows:
分类任务训练：Classification task training:
在分类任务中，任务定义如下:给定问句1,问句2，以及分类标签(其中0代表不相似，1代表相似)，通过模型需要学习到p(6|x1,x2)。In the classification task, the task is defined as follows: Given question 1, question 2, and classification labels (where 0 means dissimilar and 1 means similar), the model needs to learn p(6|x1 , x2 ).
参见图3，此时虚线内的为基础网络层，输出是模型预测的标签。See Figure 3. At this time, the basic network layer is inside the dotted line, and the output is the label predicted by the model.
排序任务训练：Sorting task training:
在排序任务中，任务定义如下：给定中心问题(记为xanchor),正样本(记为xpos),负样本(记为xneg),得到：In the sorting task, the task is defined as follows: given the central problem (denoted as xanchor ), positive samples (denoted as xpos ), and negative samples (denoted as xneg ), get:
ypos＝f(w,xanchor,xpos)ypos = f(w,xanchor ,xpos )
yneg＝f(w,xanchor,xneg)yneg ＝f(w,xanchor ,xneg )
其中f为距离函数。ypos,yneg∈(-1,1)，分别表示中心问题和正样本的相似度，以及中心问题和负样本的相似度。w是距离函数的参数。where f is the distance function. ypos , yneg ∈(-1,1), denote the similarity between the central question and the positive sample, and the similarity between the central question and the negative sample, respectively. w is the parameter of the distance function.
训练目标是：正样本和中心问题的距离尽可能的小，负样本和中心问题的距离尽可能的大，也即ypos＞yneg,排序损失函数为：The training goal is: the distance between the positive sample and the central problem is as small as possible, and the distance between the negative sample and the central problem is as large as possible, that is, ypos > yneg , and the sorting loss function is:
Lranking＝max(0,m+ypos-yneg)Lranking ＝max(0,m+ypos -yneg )
其中，Lranking为排序损失函数，m为超参数，表征正负分数之间的差距。Among them, Lranking is the ranking loss function, and m is the hyperparameter, which represents the gap between positive and negative scores.
参见图4，此时虚线内的为基础网络层，输出是排序输出网络层。Referring to Figure 4, at this time, the base network layer is inside the dotted line, and the output is the sorted output network layer.
采用多任务联合训练的方法，具体过程如下：The method of multi-task joint training is adopted, and the specific process is as follows:
联合损失函数为：The joint loss function is:
L＝Lranking+LclassificationL＝Lranking +Lclassification
其中，Lclassification为分类损失函数。因为对于分类任务和排序任务，基础网络层是共享参数的，只有输出层不同。Among them, Lclassification is the classification loss function. Because for classification tasks and sorting tasks, the basic network layer shares parameters, only the output layer is different.
所以采取交替更新的方法：在当前输入是问题分类样本时，通过Lclassification更新W和Wc，其中W是基础网络层的参数，Wc是分类输出网络层的参数；在当前输入是问题排序样本时，通过Lranking来更新W和Wr，其中Wr是排序输出网络层的参数。Therefore, an alternate update method is adopted: when the current input is a problem classification sample, update W and Wc through Lclassification , where W is the parameter of the basic network layer, and Wc is the parameter of the classification output network layer; when the current input is the problem ranking When sampling, W and Wr are updated through Lranking , where Wr is the parameter of the sorted output network layer.
本发明实施例的技术方案，采用了分类与排序联合训练的方法，来提升模型的准确率与鲁棒性。使用通用数据的分类数据集构建分类任务，旨在使得模型学习到通用的文本语义关系；同时使用目标领域数据构建了三元组数据，作为问题排序样本。The technical solution of the embodiment of the present invention adopts a classification and sorting joint training method to improve the accuracy and robustness of the model. The classification task is constructed by using the classification data set of general data, aiming to make the model learn the general text semantic relationship; at the same time, the triplet data is constructed by using the target field data as a sample of problem ranking.
对每个正样本随机采样多个负样本，使用排序损失函数进行训练，使得模型在学习到通用语义关系的同时，也能学习到相似程度的一些信息。For each positive sample, multiple negative samples are randomly sampled, and the ranking loss function is used for training, so that the model can learn some similar information while learning the general semantic relationship.
第三实施例third embodiment
图5是本申请第三实施例提供的一种问题匹配装置的结构示意图。参见图5，本实施例提供的问题匹配装置500包括：分类训练模块501和排序训练模块502。Fig. 5 is a schematic structural diagram of a question matching device provided in the third embodiment of the present application. Referring to FIG. 5 , the
其中，分类训练模块501，用于利用问题分类样本，对问题分类模型与问题排序模型复用的基础网络层，以及问题分类模型中的分类输出网络层进行训练；Wherein, the
排序训练模块502，用于利用问题排序样本，对所述基础网络层以及问题排序模型中的排序输出网络层进行训练，以得到经训练的问题排序模型供进行问题匹配。The sorting
本申请实施例的技术方案，通过在利用问题排序样本，对所述基础网络层以及问题排序模型中的排序输出网络层进行训练的基础上，利用问题分类样本，对问题分类模型与问题排序模型复用的基础网络层进行训练。因为增加了问题分类样本对基础网络层的训练，所以可以增大问题排序模型的负样本空间，从而提高问题排序模型的鲁棒性，减少问题排序模型的偏向性。In the technical solution of the embodiment of the present application, on the basis of using the problem sorting samples to train the basic network layer and the sorting output network layer in the problem sorting model, using the problem classification samples to train the problem classification model and the problem ranking model Reuse the base network layer for training. Because the training of the basic network layer with question classification samples is added, the negative sample space of the question ranking model can be increased, thereby improving the robustness of the question ranking model and reducing the bias of the question ranking model.
进一步地，在每一轮训练过程中，利用该轮问题分类样本和该轮问题排序样本分别进行训练。Further, during each round of training, the round of question classification samples and the round of question sorting samples are used for training respectively.
进一步地，在利用问题分类样本训练过程中，采用分类损失函数；Further, in the training process of using the problem classification samples, a classification loss function is used;
在利用问题排序样本训练过程中，采用排序损失函数。In the training process of ranking samples using questions, a ranking loss function is used.
进一步地，所述问题分类样本包括正样本问句对和负样本问句对；Further, the question classification samples include positive sample question pairs and negative sample question pairs;
问题排序样本包括样本三元组，其中该三元组包括中心问句、正样问句和负样问句，所述正样问句与所述中心问句的相似度大于第一相似阈值，所述负样问句与所述中心问句的相似度小于第二相似阈值，所述第一相似阈值大于或等于所述第二相似阈值。The question sorting sample includes a sample triplet, wherein the triplet includes a central question, a positive question and a negative question, and the similarity between the positive question and the central question is greater than the first similarity threshold, The similarity between the negative sample question and the central question is smaller than a second similarity threshold, and the first similarity threshold is greater than or equal to the second similarity threshold.
进一步地，所述装置还包括：Further, the device also includes:
相似度确定模块，用于得到经训练的问题排序模型之后，将用户输入的待检索问句和已有问答对中已有问句作为经训练的问题排序模型的输入，以得到所述待检索问句与所述已有问句之间的语义相似度；The similarity determination module is used to obtain the trained question ranking model, and use the user-input question sentence to be retrieved and the existing question sentence in the existing question-answer pair as the input of the trained question ranking model, so as to obtain the query sentence to be retrieved. Semantic similarity between the question and the existing question;
目标答案确定模块，用于根据所述待检索问句与所述已有问句之间的语义相似度，确定与所述待检索问句匹配的目标问句以及目标答案。The target answer determination module is configured to determine a target question sentence and a target answer matching the query sentence to be retrieved according to the semantic similarity between the query sentence to be retrieved and the existing question sentence.
进一步地，所述问题分类样本根据通用数据构建；Further, the problem classification sample is constructed according to general data;
所述问题排序样本根据目标领域数据构建。The question ranking sample is constructed according to the target field data.
第四实施例Fourth embodiment
根据本申请的实施例，本申请还提供了一种电子设备和一种可读存储介质。According to the embodiments of the present application, the present application also provides an electronic device and a readable storage medium.
如图6所示，是根据本申请实施例的问题匹配方法的电子设备的框图。电子设备旨在表示各种形式的数字计算机，诸如，膝上型计算机、台式计算机、工作台、个人数字助理、服务器、刀片式服务器、大型计算机、和其它适合的计算机。电子设备还可以表示各种形式的移动装置，诸如，个人数字处理、蜂窝电话、智能电话、可穿戴设备和其它类似的计算装置。本文所示的部件、它们的连接和关系、以及它们的功能仅仅作为示例，并且不意在限制本文中描述的和/或者要求的本申请的实现。As shown in FIG. 6 , it is a block diagram of an electronic device according to the question matching method of the embodiment of the present application. Electronic device is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are by way of example only, and are not intended to limit implementations of the applications described and/or claimed herein.
如图6所示，该电子设备包括：一个或多个处理器601、存储器602，以及用于连接各部件的接口，包括高速接口和低速接口。各个部件利用不同的总线互相连接，并且可以被安装在公共主板上或者根据需要以其它方式安装。处理器可以对在电子设备内执行的指令进行处理，包括存储在存储器中或者存储器上以在外部输入/输出装置(诸如，耦合至接口的显示设备)上显示GUI的图形信息的指令。在其它实施方式中，若需要，可以将多个处理器和/或多条总线与多个存储器和多个存储器一起使用。同样，可以连接多个电子设备，各个设备提供部分必要的操作(例如，作为服务器阵列、一组刀片式服务器、或者多处理器系统)。图6中以一个处理器601为例。As shown in FIG. 6, the electronic device includes: one or
存储器602即为本申请所提供的非瞬时计算机可读存储介质。其中，所述存储器存储有可由至少一个处理器执行的指令，以使所述至少一个处理器执行本申请所提供的问题匹配方法。本申请的非瞬时计算机可读存储介质存储计算机指令，该计算机指令用于使计算机执行本申请所提供的问题匹配方法。The
存储器602作为一种非瞬时计算机可读存储介质，可用于存储非瞬时软件程序、非瞬时计算机可执行程序以及模块，如本申请实施例中的问题匹配方法对应的程序指令/模块(例如，附图5所示的分类训练模块501和排序训练模块502)。处理器601通过运行存储在存储器602中的非瞬时软件程序、指令以及模块，从而执行服务器的各种功能应用以及数据处理，即实现上述方法实施例中的问题匹配方法。The
存储器602可以包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需要的应用程序；存储数据区可存储根据问题匹配电子设备的使用所创建的数据等。此外，存储器602可以包括高速随机存取存储器，还可以包括非瞬时存储器，例如至少一个磁盘存储器件、闪存器件、或其他非瞬时固态存储器件。在一些实施例中，存储器602可选包括相对于处理器601远程设置的存储器，这些远程存储器可以通过网络连接至问题匹配电子设备。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。The
问题匹配方法的电子设备还可以包括：输入装置603和输出装置604。处理器601、存储器602、输入装置603和输出装置604可以通过总线或者其他方式连接，图6中以通过总线连接为例。The electronic device of the question matching method may also include: an
输入装置603可接收输入的数字或字符信息，以及产生与问题匹配电子设备的用户设置以及功能控制有关的键信号输入，例如触摸屏、小键盘、鼠标、轨迹板、触摸板、指示杆、一个或者多个鼠标按钮、轨迹球、操纵杆等输入装置。输出装置604可以包括显示设备、辅助照明装置(例如，LED)和触觉反馈装置(例如，振动电机)等。该显示设备可以包括但不限于，液晶显示器(LCD)、发光二极管(LED)显示器和等离子体显示器。在一些实施方式中，显示设备可以是触摸屏。The
此处描述的系统和技术的各种实施方式可以在数字电子电路系统、集成电路系统、专用ASIC(专用集成电路)、计算机硬件、固件、软件、和/或它们的组合中实现。这些各种实施方式可以包括：实施在一个或者多个计算机程序中，该一个或者多个计算机程序可在包括至少一个可编程处理器的可编程系统上执行和/或解释，该可编程处理器可以是专用或者通用可编程处理器，可以从存储系统、至少一个输入装置、和至少一个输出装置接收数据和指令，并且将数据和指令传输至该存储系统、该至少一个输入装置、和该至少一个输出装置。Various implementations of the systems and techniques described herein can be implemented in digital electronic circuitry, integrated circuit systems, application specific ASICs (application specific integrated circuits), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include being implemented in one or more computer programs executable and/or interpreted on a programmable system including at least one programmable processor, the programmable processor Can be special-purpose or general-purpose programmable processor, can receive data and instruction from storage system, at least one input device, and at least one output device, and transmit data and instruction to this storage system, this at least one input device, and this at least one output device an output device.
这些计算程序(也称作程序、软件、软件应用、或者代码)包括可编程处理器的机器指令，并且可以利用高级过程和/或面向对象的编程语言、和/或汇编/机器语言来实施这些计算程序。如本文使用的，术语“机器可读介质”和“计算机可读介质”指的是用于将机器指令和/或数据提供给可编程处理器的任何计算机程序产品、设备、和/或装置(例如，磁盘、光盘、存储器、可编程逻辑装置(PLD))，包括，接收作为机器可读信号的机器指令的机器可读介质。术语“机器可读信号”指的是用于将机器指令和/或数据提供给可编程处理器的任何信号。These computing programs (also referred to as programs, software, software applications, or codes) include machine instructions for a programmable processor and may be implemented using high-level procedural and/or object-oriented programming languages, and/or assembly/machine language calculation program. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, apparatus, and/or means for providing machine instructions and/or data to a programmable processor ( For example, magnetic disks, optical disks, memories, programmable logic devices (PLDs), including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal used to provide machine instructions and/or data to a programmable processor.
为了提供与用户的交互，可以在计算机上实施此处描述的系统和技术，该计算机具有：用于向用户显示信息的显示装置(例如，CRT(阴极射线管)或者LCD(液晶显示器)监视器)；以及键盘和指向装置(例如，鼠标或者轨迹球)，用户可以通过该键盘和该指向装置来将输入提供给计算机。其它种类的装置还可以用于提供与用户的交互；例如，提供给用户的反馈可以是任何形式的传感反馈(例如，视觉反馈、听觉反馈、或者触觉反馈)；并且可以用任何形式(包括声输入、语音输入或者、触觉输入)来接收来自用户的输入。To provide for interaction with the user, the systems and techniques described herein can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user. ); and a keyboard and pointing device (eg, a mouse or a trackball) through which a user can provide input to the computer. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and can be in any form (including Acoustic input, speech input or, tactile input) to receive input from the user.
可以将此处描述的系统和技术实施在包括后台部件的计算系统(例如，作为数据服务器)、或者包括中间件部件的计算系统(例如，应用服务器)、或者包括前端部件的计算系统(例如，具有图形用户界面或者网络浏览器的用户计算机，用户可以通过该图形用户界面或者该网络浏览器来与此处描述的系统和技术的实施方式交互)、或者包括这种后台部件、中间件部件、或者前端部件的任何组合的计算系统中。可以通过任何形式或者介质的数字数据通信(例如，通信网络)来将系统的部件相互连接。通信网络的示例包括：局域网(LAN)、广域网(WAN)和互联网。The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., as a a user computer having a graphical user interface or web browser through which a user can interact with embodiments of the systems and techniques described herein), or including such backend components, middleware components, Or any combination of front-end components in a computing system. The components of the system can be interconnected by any form or medium of digital data communication, eg, a communication network. Examples of communication networks include: Local Area Network (LAN), Wide Area Network (WAN) and the Internet.
计算机系统可以包括客户端和服务器。客户端和服务器一般远离彼此并且通常通过通信网络进行交互。通过在相应的计算机上运行并且彼此具有客户端-服务器关系的计算机程序来产生客户端和服务器的关系。A computer system may include clients and servers. Clients and servers are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by computer programs running on the respective computers and having a client-server relationship to each other.
根据本申请实施例的技术方案，可以实现如下技术效果：According to the technical solutions of the embodiments of the present application, the following technical effects can be achieved:
1、利用通用数据构建问题分类样本，使得模型学习到了较好的通用语义信息；1. Use general data to construct problem classification samples, so that the model can learn better general semantic information;
2、利用特定领域数据构建问题排序样本，使得模型学习到了相关程度的信息，从而在重排序的过程中获得更好的结果；2. Use domain-specific data to construct problem ranking samples, so that the model can learn relevant information, so as to obtain better results in the process of reordering;
3、联合训练使得模型更加鲁棒，提高了准确率，并缓解了在特定领域数据上的过拟合问题。3. Joint training makes the model more robust, improves the accuracy rate, and alleviates the problem of over-fitting on data in specific fields.
应该理解，可以使用上面所示的各种形式的流程，重新排序、增加或删除步骤。例如，本发申请中记载的各步骤可以并行地执行也可以顺序地执行也可以不同的次序执行，只要能够实现本申请公开的技术方案所期望的结果，本文在此不进行限制。It should be understood that steps may be reordered, added or deleted using the various forms of flow shown above. For example, the steps described in the present application may be executed in parallel, sequentially, or in a different order, as long as the desired result of the technical solution disclosed in the present application can be achieved, no limitation is imposed herein.
上述具体实施方式，并不构成对本申请保护范围的限制。本领域技术人员应该明白的是，根据设计要求和其他因素，可以进行各种修改、组合、子组合和替代。任何在本申请的精神和原则之内所作的修改、等同替换和改进等，均应包含在本申请保护范围之内。The above specific implementation methods are not intended to limit the protection scope of the present application. It should be apparent to those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made depending on design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of this application shall be included within the protection scope of this application.
Claims (10)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910843848.3ACN110543558B (en) | 2019-09-06 | 2019-09-06 | Problem matching method, device, equipment and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910843848.3ACN110543558B (en) | 2019-09-06 | 2019-09-06 | Problem matching method, device, equipment and medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110543558A CN110543558A (en) | 2019-12-06 |
| CN110543558Btrue CN110543558B (en) | 2023-01-17 |
Family
ID=68712875
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910843848.3AActiveCN110543558B (en) | 2019-09-06 | 2019-09-06 | Problem matching method, device, equipment and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110543558B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111709247B (en)* | 2020-05-20 | 2023-04-07 | 北京百度网讯科技有限公司 | Data set processing method and device, electronic equipment and storage medium |
| CN113836885B (en)* | 2020-06-24 | 2025-02-25 | 阿里巴巴集团控股有限公司 | Text matching model training method, text matching method, device and electronic device |
| CN112328710B (en)* | 2020-11-26 | 2024-06-11 | 北京百度网讯科技有限公司 | Entity information processing method, device, electronic equipment and storage medium |
| CN113254609B (en)* | 2021-05-12 | 2022-08-09 | 同济大学 | Question-answering model integration method based on negative sample diversity |
| CN113434792B (en)* | 2021-07-20 | 2023-07-18 | 北京百度网讯科技有限公司 | Training method of network address matching model and network address matching method |
| CN114003688B (en)* | 2021-10-14 | 2025-07-29 | 咪咕文化科技有限公司 | Query method, device, equipment and storage medium for question and answer data |
| CN114119989B (en)* | 2021-11-29 | 2023-08-11 | 北京百度网讯科技有限公司 | Image feature extraction model training method, device and electronic equipment |
| CN114357132A (en)* | 2021-12-27 | 2022-04-15 | 北京百度网讯科技有限公司 | Generating method, sorting method, apparatus, device and storage medium of sorting model |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7991764B2 (en)* | 2005-07-22 | 2011-08-02 | Yogesh Chunilal Rathod | Method and system for communication, publishing, searching, sharing and dynamically providing a journal feed |
| CN105893523B (en)* | 2016-03-31 | 2019-05-17 | 华东师范大学 | It is measured using the assessment of answer relevance ranking come the method for computational problem similarity |
| EP3454260A1 (en)* | 2017-09-11 | 2019-03-13 | Tata Consultancy Services Limited | Bilstm-siamese network based classifier for identifying target class of queries and providing responses thereof |
| CN109948143B (en)* | 2019-01-25 | 2023-04-07 | 网经科技（苏州）有限公司 | Answer extraction method of community question-answering system |
- 2019
- 2019-09-06CNCN201910843848.3Apatent/CN110543558B/enactiveActive
Also Published As
| Publication number | Publication date |
|---|---|
| CN110543558A (en) | 2019-12-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110543558B (en) | Problem matching method, device, equipment and medium | |
| CN112560912B (en) | Classification model training methods, devices, electronic equipment and storage media | |
| EP3828719A2 (en) | Method and apparatus for generating model for representing heterogeneous graph node, electronic device, storage medium, and computer program product | |
| CN111639710B (en) | Image recognition model training method, device, equipment and storage medium | |
| CN111125335B (en) | Question and answer processing method, device, electronic device and storage medium | |
| CN111522967B (en) | Knowledge graph construction method, device, equipment and storage medium | |
| CN111325020B (en) | Event argument extraction method and device and electronic equipment | |
| KR102645185B1 (en) | Method, apparatus, electronic device, program and readable storage medium for creating a label marking model | |
| KR102475235B1 (en) | Method for resource sorting, method for training sorting model and corresponding apparatuses | |
| EP3923160A1 (en) | Method, apparatus, device and storage medium for training model | |
| CN110795569B (en) | Method, device and device for generating vector representation of knowledge graph | |
| CN111104514B (en) | Training method and device for document tag model | |
| US20210390260A1 (en) | Method, apparatus, device and storage medium for matching semantics | |
| CN111950291B (en) | Semantic representation model generation method and device, electronic equipment and storage medium | |
| CN110705460B (en) | Image category recognition method and device | |
| CN111221984A (en) | Multimodal content processing method, device, device and storage medium | |
| CN112001180A (en) | Multi-mode pre-training model acquisition method and device, electronic equipment and storage medium | |
| CN111681647B (en) | Method, device, equipment and storage medium for identifying word slots | |
| CN114860913B (en) | Intelligent question and answer system construction method, question and answer processing method and device | |
| CN111241234B (en) | Text classification method and device | |
| CN111522944B (en) | Method, apparatus, device and storage medium for outputting information | |
| CN111709252B (en) | Model improvement method and device based on pre-trained semantic model | |
| CN111324715A (en) | Method and device for generating question-answering robot | |
| CN111813828B (en) | Entity relation mining method and device, electronic equipment and storage medium | |
| CN111950293B (en) | Semantic representation model generation method and device, electronic equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |