技术领域Technical Field
本申请涉及自然语言处理技术领域，具体而言，涉及一种图文评级方法、装置及计算机可读存储。The present application relates to the technical field of natural language processing, and in particular to a method and device for rating pictures and texts, and computer-readable storage.
背景技术Background technique
随着信息技术的发展，各种领域的图文信息应运而生，例如：新闻、资讯、自媒体等发布的图文信息及公司的研究报告等，为了便于用户有针对性地、快速地获取信息，可以对图文信息进行分类。With the development of information technology, graphic and text information in various fields has emerged, such as news, information, graphic and text information published by self-media, and company research reports. In order to facilitate users to obtain information in a targeted and rapid manner, graphic and text information can be classified.
在现有技术中，通常采用多模态信息对图文内容进行分类的方法，而该方法或是需要人工标记的多模态分类标签训练多模态模型，或是需要大规模的高质量图像-文本数据对训练多模态模型。其中，采用人工标记的多模态分类标签训练多模态模型的方式不仅耗费大量的人力资源，而且在人工标记的过程中可能存在主观性和一致性的问题，从而影响模型的性能，而采用大规模的高质量图像-文本数据对训练多模态模型，由于描述文本中的关键词一般为图像中的元素，因此模型的性能非常依赖标注的准确性，同时，还可能由于关键词无法概述图像内容而导致语义鸿沟，影响模型的表现。In the prior art, a method of classifying graphic content using multimodal information is usually used, and this method either requires manually labeled multimodal classification labels to train the multimodal model, or requires large-scale high-quality image-text data pairs to train the multimodal model. Among them, the method of training the multimodal model using manually labeled multimodal classification labels not only consumes a lot of human resources, but also may have subjective and consistency problems in the manual labeling process, thereby affecting the performance of the model. The method of training the multimodal model using large-scale high-quality image-text data pairs, because the keywords in the description text are generally elements in the image, the performance of the model is very dependent on the accuracy of the annotation. At the same time, it may also cause a semantic gap due to the inability of keywords to summarize the image content, affecting the performance of the model.
发明内容Summary of the invention
为了至少克服现有技术中的上述不足，本申请的目的在于提供一种图文评级方法、装置及计算机可读存储。In order to at least overcome the above-mentioned deficiencies in the prior art, the purpose of the present application is to provide a method, device and computer-readable storage for rating pictures and texts.
第一方面，本申请实施例提供一种图文评级方法，所述图文评级方法包括：In a first aspect, an embodiment of the present application provides a method for rating pictures and texts, the method comprising:
获取待评级的图文内容，所述图文内容包括图片内容和文字内容；Obtaining graphic content to be rated, wherein the graphic content includes picture content and text content;
检测所述文字内容中的关键词信息，并基于所述关键词信息对所述文字内容进行评分，得到文字内容评分；Detecting keyword information in the text content, and scoring the text content based on the keyword information to obtain a text content score;
基于所述图片内容的图像质量对所述图片内容进行评分，得到所述图片内容评分；Scoring the picture content based on the image quality of the picture content to obtain the picture content score;
基于所述图片内容与所述文字内容的匹配程度进行评分，得到内容匹配度评分；Scoring the image content based on the degree of match between the image content and the text content to obtain a content match score;
基于所述文字内容评分、所述图片内容评分及所述内容匹配度评分对所述图文内容进行评级。The graphic content is rated based on the text content score, the picture content score and the content matching score.
在一种可能的实现方式中，所述检测所述文字内容中的关键词信息，并基于所述关键词信息对所述文字内容进行评分，得到文字内容评分的步骤之前，所述方法还包括构建关键词知识库的步骤，该步骤包括：In a possible implementation, before the step of detecting keyword information in the text content and scoring the text content based on the keyword information to obtain a text content score, the method further includes the step of constructing a keyword knowledge base, which includes:
收集目标场景中的语料数据及所述语料数据对应的品类信息；Collecting corpus data in the target scenario and category information corresponding to the corpus data;
对所述语料数据进行分词统计，得到语料关键词；Perform word segmentation statistics on the corpus data to obtain corpus keywords;
计算所述语料关键词与各个所述品类信息之间的关联值，所述关联值用于表征所述语料关键词和各个所述品类信息之间的关联度；Calculating the association value between the corpus keyword and each of the category information, wherein the association value is used to represent the degree of association between the corpus keyword and each of the category information;
基于所述品类信息、所述语料关键词及所述关联值构建所述关键词知识库。The keyword knowledge base is constructed based on the category information, the corpus keywords and the associated values.
在一种可能的实现方式中，所述检测所述文字内容中的关键词信息，并基于所述关键词信息对所述文字内容进行评分，得到文字内容评分的步骤，包括：In a possible implementation, the step of detecting keyword information in the text content, and scoring the text content based on the keyword information to obtain a text content score includes:
获取所述文字内容的关键词信息及对应的第一品类信息；Obtain keyword information of the text content and corresponding first category information;
根据所述关键词信息得到对应的命名实体类型，并根据所述命名实体类型的数量及频次对所述文字内容进行评分，得到内容评分；Obtaining corresponding named entity types according to the keyword information, and scoring the text content according to the number and frequency of the named entity types to obtain a content score;
基于所述第一品类信息和所述关键词信息对所述文字内容进行评分，得到主题评分；Scoring the text content based on the first category information and the keyword information to obtain a topic score;
识别所述关键词信息中的情绪关键词，根据所述情绪关键词对所述文字内容的情绪类型进行分类，并基于所述文字内容的情绪类型进行评分，得到情绪评分；Identifying emotional keywords in the keyword information, classifying the emotional types of the text content according to the emotional keywords, and scoring based on the emotional types of the text content to obtain an emotional score;
判断所述文字内容的通顺度，并基于所述通顺度进行评分得到通顺度评分；Determining the fluency of the text content, and scoring based on the fluency to obtain a fluency score;
基于所述内容评分、所述主题评分、所述情绪评分及所述通顺度评分得到所述文字内容评分。The text content score is obtained based on the content score, the topic score, the sentiment score and the fluency score.
在一种可能的实现方式中，所述基于所述第一品类信息和所述关键词信息对所述文字内容进行评分，得到主题评分的步骤，包括：In a possible implementation, the step of scoring the text content based on the first category information and the keyword information to obtain a topic score includes:
基于所述关键词知识库得到所述第一品类信息与所述关键词信息对应的关联值，将数值最大的关联值作为所述主题评分。Based on the keyword knowledge base, the correlation value corresponding to the first category information and the keyword information is obtained, and the correlation value with the largest value is used as the topic score.
在一种可能的实现方式中，所述判断所述文字内容的通顺度，并基于所述通顺度进行评分得到所述通顺度评分的步骤，包括：In a possible implementation, the step of determining the fluency of the text content and scoring based on the fluency to obtain the fluency score includes:
按所述文字内容中词语顺序依次获取待预测的词语，基于所述待预测的词语之前的词语对所述待预测的词语进行猜词预测，根据所述待预测的词语在预测结果中的出现概率评估所述文字内容的通顺度，并基于所述通顺度进行评分得到所述通顺度评分。The words to be predicted are obtained in sequence according to the order of words in the text content, the words to be predicted are guessed and predicted based on the words before the words to be predicted, the fluency of the text content is evaluated according to the probability of occurrence of the words to be predicted in the prediction results, and a score is scored based on the fluency to obtain the fluency score.
在一种可能的实现方式中，所述基于所述图片内容与所述文字内容的匹配程度进行评分，得到内容匹配度评分的步骤，包括：In a possible implementation, the step of scoring based on the degree of matching between the image content and the text content to obtain a content matching score includes:
识别所述图片内容，并基于识别结果生成相应的标签；Identify the image content and generate corresponding tags based on the identification results;
基于所述关键词知识库得到各个所述标签对应的第二品类信息，并基于所述关键词知识库得到各个所述关键词信息对应的第三品类信息；Based on the keyword knowledge base, obtain the second category information corresponding to each of the tags, and based on the keyword knowledge base, obtain the third category information corresponding to each of the keyword information;
计算所述第二品类信息及所述第三品类信息的分布差异，基于所述分布差异计算所述图片内容与所述文字内容的匹配程度，得到匹配度分数。The distribution difference between the second category information and the third category information is calculated, and the matching degree between the image content and the text content is calculated based on the distribution difference to obtain a matching score.
在一种可能的实现方式中，所述基于所述文字内容评分、所述图片内容评分及所述内容匹配度评分对所述图文内容进行评级的步骤，包括：In a possible implementation, the step of rating the graphic content based on the text content score, the picture content score, and the content matching score includes:
对所述文字内容评分、所述图片内容评分及所述内容匹配度评分进行归一化处理，将所述文字内容评分、所述图片内容评分及所述内容匹配度评分映射到预设的分数范围；Normalizing the text content score, the picture content score, and the content matching score, and mapping the text content score, the picture content score, and the content matching score to a preset score range;
基于归一化处理后的文字内容评分、图片内容评分及内容匹配度评分得到一加权分数；A weighted score is obtained based on the normalized text content score, image content score and content matching score;
检测所述图文内容是否存在广告信息，当所述图文内容不存在广告信息时，将所述加权分数作为最终的评级分数，当所述图文内容存在广告信息时，对所述加权分数进行降分处理，将降分处理后的加权分数作为最终的评级分数；Detecting whether there is advertising information in the graphic content, and when there is no advertising information in the graphic content, using the weighted score as the final rating score; when there is advertising information in the graphic content, downgrading the weighted score, and using the weighted score after downgrading as the final rating score;
基于所述评级分数对所述图文内容进行评级，得到所述图文内容的评级结果。The graphic content is rated based on the rating score to obtain a rating result of the graphic content.
在一种可能的实现方式中，所述基于所述评级分数对所述图文内容进行评级，得到所述图文内容的评级结果的步骤之后，所述方法还包括：In a possible implementation, after the step of rating the graphic content based on the rating score to obtain a rating result of the graphic content, the method further includes:
基于所述评级结果得到所述图文内容对应的推荐指数，并基于所述推荐指数对所述图文内容进行推荐；Obtaining a recommendation index corresponding to the graphic content based on the rating result, and recommending the graphic content based on the recommendation index;
基于所述评级结果及所述评级分数生成对应的评级报告，并将所述评级报告发送至客户端。A corresponding rating report is generated based on the rating result and the rating score, and the rating report is sent to the client.
第二方面，本申请实施例还提供一种图文评级装置，所述装置包括：In a second aspect, an embodiment of the present application further provides a picture and text rating device, the device comprising:
获取模块，用于获取待评级的图文内容，所述图文内容包括图片内容和文字内容；An acquisition module, used to acquire the graphic content to be rated, wherein the graphic content includes picture content and text content;
第一评分模块，用于检测所述文字内容中的关键词信息，并基于所述关键词信息对所述文字内容进行评分，得到文字内容评分；A first scoring module is used to detect keyword information in the text content and score the text content based on the keyword information to obtain a text content score;
第二评分模块，用于基于所述图片内容的图像质量对所述图片内容进行评分，得到所述图片内容评分；A second scoring module is used to score the picture content based on the image quality of the picture content to obtain the picture content score;
第三评分模块，用于基于所述图片内容与所述文字内容的匹配程度进行评分，得到内容匹配度评分；A third scoring module is used to score based on the matching degree between the image content and the text content to obtain a content matching score;
评级模块，用于基于所述文字内容评分、所述图片内容评分及所述内容匹配度评分对所述图文内容进行评级。The rating module is used to rate the graphic content based on the text content rating, the image content rating and the content matching rating.
第三方面，本申请实施例还提供一种计算机可读存储介质，其上存储有计算机程序，该计算机程序被处理器执行时实现如第一方面中任意一项所述的图文评级方法。In a third aspect, an embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the image and text rating method as described in any one of the first aspects.
基于上述任意一个方面，本申请实施例提供的图文评级方法、装置及计算机可读存储，通过对文字内容、图片内容和图文内容的匹配度等进行评分，并基于文字内容评分、图片内容评分和内容匹配度评分对图文内容进行评级，从而可以综合不同模态的信息全面、综合地评估图文质量，提高评级结果的可信度及准确度。此外，本方案无需多模态标签数据和高质量图像-文本数据对训练模型，且在实际预测过程中并不约束文本中的关键词一定出现于图像中，不仅可以节约人力资源，还可以提高模型的性能，提高评级结果的可信度及准确度。Based on any of the above aspects, the text and image rating method, device and computer-readable storage provided in the embodiments of the present application can comprehensively and comprehensively evaluate the quality of text and image by scoring the matching degree of text content, image content and text and image content, and rating the text and image content based on the text content score, image content score and content matching score, thereby improving the credibility and accuracy of the rating results by integrating information from different modalities. In addition, this solution does not require multimodal label data and high-quality image-text data to train the model, and in the actual prediction process, it does not constrain the keywords in the text to appear in the image, which can not only save human resources, but also improve the performance of the model and the credibility and accuracy of the rating results.
附图说明BRIEF DESCRIPTION OF THE DRAWINGS
为了更清楚地说明本申请实施例的技术方案，下面将对实施例中所需要调用的附图作简单地介绍，应当理解，以下附图仅示出了本申请的某些实施例，因此不应被看作是对范围的限定，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其它相关的附图。In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.
图1为本申请实施例提供的图文评级方法的一种流程示意图；FIG1 is a schematic diagram of a flow chart of a method for rating pictures and texts provided in an embodiment of the present application;
图2为图1中可能包括的一种子步骤流程示意图；FIG2 is a schematic diagram of a sub-step process that may be included in FIG1;
图3为图1中步骤S120的子步骤流程示意图；FIG3 is a schematic diagram of the sub-step flow chart of step S120 in FIG1 ;
图4为图1中步骤S140的子步骤流程示意图；FIG4 is a schematic diagram of the sub-step flow chart of step S140 in FIG1 ;
图5为图1中步骤S150的子步骤流程示意图；FIG5 is a schematic diagram of the sub-step flow chart of step S150 in FIG1 ;
图6为图1中可能包括的另一种子步骤流程示意图；FIG6 is a schematic diagram of another seed step process that may be included in FIG1;
图7为本申请实施例提供的图文评级装置的功能模块图；FIG7 is a functional module diagram of a picture and text rating device provided in an embodiment of the present application;
图8为本申请实施例提供的服务器的结构方框示意图。FIG8 is a schematic block diagram of the structure of the server provided in an embodiment of the present application.
具体实施方式Detailed ways
为使本申请实施例的目的、技术方案和优点更加清楚，下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，应当理解，本申请中附图仅起到说明和描述的目的，并不用于限定本申请的保护范围。另外，应当理解，示意性的附图并未按实物比例绘制。本申请中使用的流程图示出了根据本申请实施例的一些实施例实现的操作。应该理解，流程图的操作可以不按顺序实现，没有逻辑的上下文关系的步骤可以反转顺序或者同时实施。此外，本领域技术人员在本申请内容的指引下，可以向流程图添加一个或多个其它操作，也可以从流程图中移除一个或多个操作。To make the purpose, technical scheme and advantages of the embodiment of the present application clearer, the technical scheme in the embodiment of the present application will be clearly and completely described below in conjunction with the drawings in the embodiment of the present application. It should be understood that the drawings in the present application only serve the purpose of explanation and description and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn in real proportion. The flowchart used in this application shows the operations implemented according to some embodiments of the embodiment of the present application. It should be understood that the operations of the flowchart can be implemented out of order, and the steps without logical context can be reversed in order or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart under the guidance of the content of the present application, or remove one or more operations from the flowchart.
另外，所描述的实施例仅仅是本申请一部分实施例，而不是全部的实施例。通常在此处附图中描述和示出的本申请实施例的组件可以以各种不同的配置来布置和设计。因此，以下对在附图中提供的本申请的实施例的详细描述并非旨在限制要求保护的本申请的范围，而是仅仅表示本申请的选定实施例。基于本申请的实施例，本领域技术人员在没有做出创造性劳动的前提下所获得的所有其它实施例，都属于本申请保护的范围。In addition, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.
需要说明的是，在不冲突的情况下，本申请的实施例中的不同特征之间可以相互结合。It should be noted that, in the absence of conflict, different features in the embodiments of the present application may be combined with each other.
下面结合图1对本申请实施例提供的图文评级方法进行示例性说明。请参阅图1，本申请实施例提供的图文评级方法可以由服务器执行，在本申请实施例的图文评级方法中的部分步骤的顺序可以根据实际需要相互交换，或者其中的部分步骤也可以省略或删除，该服务器执行的图文评级方法的详细步骤介绍如下。The following is an exemplary description of the image and text rating method provided in the embodiment of the present application in conjunction with Figure 1. Referring to Figure 1, the image and text rating method provided in the embodiment of the present application can be executed by a server, and the order of some steps in the image and text rating method in the embodiment of the present application can be interchanged according to actual needs, or some steps can be omitted or deleted. The detailed steps of the image and text rating method executed by the server are described as follows.
步骤S110：获取待评级的图文内容，图文内容包括图片内容和文字内容。Step S110: obtaining the graphic content to be rated, where the graphic content includes picture content and text content.
在本步骤中，待评级的图文内容包括，但不限于新闻、资讯、自媒体等发布的图文信息及公司的研究报告等，文字内容可以包括标题文本、正文文本等。In this step, the graphic and textual content to be rated includes, but is not limited to, graphic and textual information published by news, information, self-media, etc., and company research reports, etc., and the textual content may include title text, body text, etc.
步骤S120：检测文字内容中的关键词信息，并基于关键词信息对文字内容进行评分，得到文字内容评分。Step S120: Detect keyword information in the text content, and score the text content based on the keyword information to obtain a text content score.
在本步骤中，通过关键词信息可以从文字的表达、主题及情绪等维度对文字内容进行评估，从而提高评级结果的可靠性。In this step, keyword information can be used to evaluate the text content from the dimensions of text expression, theme and emotion, thereby improving the reliability of the rating results.
步骤S130：基于图片内容的图像质量对图片内容进行评分，得到图片内容评分。Step S130: Score the picture content based on the image quality of the picture content to obtain a picture content score.
在本步骤中，由于图片内容的质量会影响图文内容的整体质量，故而可以基于美学质量和清晰度等维度对图片内容进行质量评估，示例性地，可以利用AVA美学评估数据集及KonIQ-10K数据集对深度学习模型训练得到的模型对图片内容进行评分，最终的图片内容评分可以是不同维度的评价分数的加权分数。In this step, since the quality of the image content will affect the overall quality of the graphic content, the image content quality can be evaluated based on dimensions such as aesthetic quality and clarity. For example, the model trained by the deep learning model using the AVA aesthetic evaluation dataset and the KonIQ-10K dataset can be used to score the image content. The final image content score can be a weighted score of the evaluation scores of different dimensions.
步骤S140：基于图片内容与文字内容的匹配程度进行评分，得到内容匹配度评分。Step S140: Score the image content and the text content based on their matching degree to obtain a content matching score.
在本步骤中，通过考量文字内容与图片内容的匹配程度，可以全面评估图文内容的一致性和协调性，从而更为全面、准确地评估整体的图文质量，提高评级结果的可信度。In this step, by considering the degree of match between text content and image content, the consistency and coordination of the image and text content can be comprehensively evaluated, thereby more comprehensively and accurately evaluating the overall image and text quality and improving the credibility of the rating results.
步骤S150：基于文字内容评分、图片内容评分及内容匹配度评分对图文内容进行评级。Step S150: Rating the graphic content based on the text content score, the image content score and the content matching score.
在本实施例中，基于文字内容评分、图片内容评分和内容匹配度评分对图文内容进行评级，可以综合不同模态的信息全面、综合地评估图文质量，使得评级结果更具可信度及准确度。In this embodiment, the graphic content is rated based on the text content rating, the image content rating and the content matching rating, which can comprehensively and integratedly evaluate the quality of the graphic content by integrating information from different modalities, making the rating results more credible and accurate.
在本实施例中，通过对文字内容、图片内容和图文内容的匹配度等进行评分，并基于文字内容评分、图片内容评分和内容匹配度评分对图文内容进行评级，从而可以综合不同模态的信息全面、综合地评估图文质量，提高评级结果的可信度及准确度。此外，本方案无需多模态标签数据和高质量图像-文本数据对训练模型，且在实际预测过程中并不约束文本中的关键词一定出现于图像中，不仅可以节约人力资源，还可以提高模型的性能，提高评级结果的可信度及准确度。In this embodiment, by scoring the text content, image content, and the matching degree of the image and text content, and rating the image and text content based on the text content score, image content score, and content matching score, the image and text quality can be comprehensively and comprehensively evaluated by integrating information from different modalities, thereby improving the credibility and accuracy of the rating results. In addition, this solution does not require multimodal label data and high-quality image-text data to train the model, and in the actual prediction process, it does not constrain the keywords in the text to appear in the image, which can not only save human resources, but also improve the performance of the model and improve the credibility and accuracy of the rating results.
作为一种可能的实现方式，请参考图2，本申请实施例的步骤S120之前还可以包括构建关键词知识库的步骤，该步骤可以通过以下方法实现。As a possible implementation, please refer to FIG. 2 . The embodiment of the present application may further include a step of constructing a keyword knowledge base before step S120 . This step may be implemented by the following method.
步骤S101：收集目标场景中的语料数据及语料数据对应的品类信息。Step S101: Collect corpus data in a target scenario and category information corresponding to the corpus data.
在本步骤中，语料数据可以包括弹幕、标题、评论等数据，这些语料数据通常具有对应的品类信息，具体地，品类信息可以由用户上传时进行选择得到，也可以经由运营人员和/或深度模型基于上传内容进行分类。In this step, the corpus data may include barrage, title, comment and other data, which usually have corresponding category information. Specifically, the category information can be obtained by user selection when uploading, or can be classified by operators and/or deep models based on the uploaded content.
步骤S102：对语料数据进行分词统计，得到语料关键词。Step S102: Perform word segmentation statistics on the corpus data to obtain corpus keywords.
在本步骤中，可以先对语料数据进行预处理(例如，去除“的”、“是”等停用词及长尾词等)，再对语料数据进行分词统计，根据统计结果得到语料关键词，示例性地，可以根据词语的出现频率确定语料关键词。In this step, the corpus data can be preprocessed first (for example, removing stop words and long-tail words such as "的" and "是"), and then the corpus data can be segmented and counted, and the corpus keywords can be obtained based on the statistical results. For example, the corpus keywords can be determined based on the frequency of occurrence of the words.
步骤S103：计算语料关键词与各个品类信息之间的关联值，关联值用于表征语料关键词和各个品类信息之间的关联度。Step S103: calculating the association value between the corpus keyword and each category information, where the association value is used to represent the association degree between the corpus keyword and each category information.
在本步骤中，可以将每一品类下的全部语料数据视为一品类文档，关联值可以是词频-逆文档频率(TF-IDF)，具体地，某个关键词在某个品类文档中的词频-逆文档频率(TF-IDF)计算方式如下：In this step, all corpus data under each category can be regarded as a category document, and the associated value can be term frequency-inverse document frequency (TF-IDF). Specifically, the term frequency-inverse document frequency (TF-IDF) of a keyword in a category document is calculated as follows:
TF-IDF＝TF×IDF；TF-IDF=TF×IDF;
其中，词频TF(Term Frequency)用于表征词语在某品类文档中的出现频率，逆文档频率IDF(Inverse Document Frequency)用于表征词语的普遍程度，当包含该词语的品类文档数量越少，则该词语区别品类文档的能力越强。Among them, term frequency TF (Term Frequency) is used to characterize the frequency of occurrence of a word in a certain category of documents, and inverse document frequency IDF (Inverse Document Frequency) is used to characterize the prevalence of a word. The fewer the number of category documents containing the word, the stronger the ability of the word to distinguish category documents.
步骤S104：基于品类信息、语料关键词及关联值构建关键词知识库。Step S104: construct a keyword knowledge base based on category information, corpus keywords and associated values.
在本步骤中，通过查询关键词知识库快速得知关键词与品类信息的关联程度，便于实施后续的图文评分步骤。此外，可以定期根据新的数据及用户的反馈数据优化关键词知识库，以保证语料关键词与品类信息关联的准确性和时效性。In this step, the degree of association between keywords and category information is quickly obtained by querying the keyword knowledge base, which facilitates the implementation of the subsequent image and text scoring step. In addition, the keyword knowledge base can be regularly optimized based on new data and user feedback data to ensure the accuracy and timeliness of the association between corpus keywords and category information.
在本实施例中，基于品类信息、语料关键词及关联值构建关键词知识库，可以通过查询关键词知识库快速得知关键词与品类信息的关联程度，便于实施后续的图文评分步骤。In this embodiment, a keyword knowledge base is constructed based on category information, corpus keywords and associated values. The degree of association between keywords and category information can be quickly learned by querying the keyword knowledge base, which facilitates the implementation of subsequent image and text scoring steps.
作为一种可能的实现方式，请参考图3，本申请实施例的步骤S120可以通过以下方式实现。As a possible implementation method, please refer to FIG. 3 , step S120 of the embodiment of the present application can be implemented in the following manner.
步骤S121：获取文字内容的关键词信息及对应的第一品类信息。Step S121: Obtain keyword information of the text content and corresponding first category information.
在本步骤中，可以先对文字内容进行预处理(例如，去除“的”、“是”等停用词及长尾词等)，再对文字内容进行分词统计，根据统计结果得到关键词信息，示例性地，可以根据词语的出现频率确定关键词信息。品类信息可以由用户上传时进行选择得到，也可以经由运营人员和/或深度模型基于上传内容进行分类。In this step, the text content can be preprocessed (for example, stop words such as "的" and "是" and long-tail words, etc. are removed), and then the text content is segmented and counted, and keyword information is obtained according to the statistical results. For example, the keyword information can be determined according to the frequency of occurrence of the words. Category information can be obtained by the user when uploading, or it can be classified by operators and/or deep models based on the uploaded content.
步骤S122：根据关键词信息得到对应的命名实体类型，并根据命名实体类型的数量及频次对文字内容进行评分，得到内容评分。Step S122: Obtain corresponding named entity types according to the keyword information, and score the text content according to the number and frequency of the named entity types to obtain a content score.
在本步骤中，内容评分可以用于评估文字内容中的看点是否丰富，具体地，可以通过文本解析工具(例如，TexSmart等)对文字内容进行解析，得到文字内容中的命名实体(Named Entity)，同时启发式地设定具备高看点的命名实体类型(例如，人物、产品、作品、事件、组织及角色等)，同时基于命名实体类型的数量及频次(注意，同一命名实体类型仅记录一次)对文字内容的看点进行打分，当命名实体类型的数量越多，且频次越高的情况下，可以认为该文字内容的看点更丰富，示例性地，打分规则可以是：score＝min(1,0.2+0.4×num(entity))。In this step, the content score can be used to evaluate whether the text content is rich in highlights. Specifically, the text content can be parsed by a text parsing tool (for example, TexSmart, etc.) to obtain the named entities in the text content, and at the same time, the named entity types with high highlights (for example, people, products, works, events, organizations and roles, etc.) are heuristically set. At the same time, the highlights of the text content are scored based on the number and frequency of the named entity types (note that the same named entity type is only recorded once). When the number of named entity types is greater and the frequency is higher, it can be considered that the text content has richer highlights. Exemplarily, the scoring rule can be: score = min (1, 0.2 + 0.4 × num (entity)).
步骤S123：基于第一品类信息和关键词信息对文字内容进行评分，得到主题评分。Step S123: Score the text content based on the first category information and the keyword information to obtain a topic score.
在本步骤中，可以基于关键词知识库得到第一品类信息与关键词信息对应的关联值，将数值最大的关联值作为主题评分，该主题评分可以表征文字内容的主题明确度。In this step, the correlation value corresponding to the first category information and the keyword information can be obtained based on the keyword knowledge base, and the correlation value with the largest value is used as the topic score, which can represent the topic clarity of the text content.
步骤S124：识别关键词信息中的情绪关键词，根据情绪关键词对文字内容的情绪类型进行分类，并基于文字内容的情绪类型进行评分，得到情绪评分。Step S124: identifying emotional keywords in the keyword information, classifying the emotional types of the text content according to the emotional keywords, and scoring based on the emotional types of the text content to obtain an emotional score.
在本步骤中，优质的文字内容可以唤醒观众的情绪，提高用户的阅读快感，因而通过识别情绪关键词对文字内容的情绪进行分类、评分，可以基于情绪评分评估文字内容对观众情绪的感染力。In this step, high-quality text content can awaken the audience's emotions and improve the user's reading pleasure. Therefore, by identifying emotional keywords to classify and score the emotions of the text content, the appeal of the text content to the audience's emotions can be evaluated based on the emotional score.
步骤S125：判断文字内容的通顺度，并基于通顺度进行评分得到通顺度评分。Step S125: determining the fluency of the text content, and scoring based on the fluency to obtain a fluency score.
在本步骤中，可以按文字内容中词语顺序依次获取待预测的词语，基于待预测的词语之前的词语对待预测的词语进行猜词预测，根据待预测的词语在预测结果中的出现概率评估文字内容的通顺度，并基于通顺度进行评分得到通顺度评分，通过通顺度评分可以评估文字内容的语句是否通顺。In this step, the words to be predicted can be obtained in sequence according to the order of words in the text content, and the words to be predicted can be guessed based on the words before the words to be predicted. The fluency of the text content is evaluated according to the probability of occurrence of the words to be predicted in the prediction results, and a score is given based on the fluency to obtain a fluency score. The fluency of the text content can be evaluated by the fluency score.
步骤S126：基于内容评分、主题评分、情绪评分及通顺度评分得到文字内容评分。Step S126: Obtaining a text content score based on the content score, theme score, sentiment score, and fluency score.
在本步骤中，基于内容评分、主题评分、情绪评分及通顺度评分评分得到的文字内容评分，可以从看点、主题明确度、情绪调动性及语句通顺度等维度全面评估文字内容的质量，从而提高评级结果的可信度。In this step, the text content score obtained based on the content score, theme score, emotion score and fluency score can comprehensively evaluate the quality of the text content from the dimensions of highlights, theme clarity, emotion mobilization and sentence fluency, thereby improving the credibility of the rating results.
在本步骤中，从看点、主题明确度、情绪调动性及语句通顺度等维度对文字内容进行评分，再基于内容评分、主题评分、情绪评分及通顺度评分评分得到的文字内容评分，可以全面、综合地评估文字内容的质量，从而提高评级结果的可信度。In this step, the text content is scored based on the dimensions of highlights, theme clarity, emotional mobilization and sentence fluency. The text content score obtained based on the content score, theme score, emotional score and fluency score can comprehensively and comprehensively evaluate the quality of the text content, thereby improving the credibility of the rating results.
进一步地，请参考图4，步骤S140可以通过以下方法实现。Further, referring to FIG. 4 , step S140 may be implemented by the following method.
步骤S141：识别图片内容，并基于识别结果生成相应的标签。Step S141: Identify the image content and generate corresponding tags based on the identification results.
步骤S142：基于关键词知识库得到各个标签对应的第二品类信息，并基于关键词知识库得到各个关键词信息对应的第三品类信息。Step S142: obtaining the second category information corresponding to each tag based on the keyword knowledge base, and obtaining the third category information corresponding to each keyword information based on the keyword knowledge base.
步骤S143：计算第二品类信息及第三品类信息的分布差异，基于分布差异计算图片内容与文字内容的匹配程度，得到匹配度分数。Step S143: Calculate the distribution difference between the second category information and the third category information, and calculate the matching degree between the image content and the text content based on the distribution difference to obtain a matching score.
在本实施例中，可以通过视觉模型识别图片中的目标对象，得到目标对象对应的标签，该标签可以视作图像的关键词信息。此外，可以通过查询关键词知识库得到对应的第二品类信息及第三品类信息，并采用JS散度(Jensen-Shannon Divergence)来刻画两组品类信息之间的分布差异，基于两两之间的分布差异的最小值计算图片内容与文字内容的匹配程度，其中，JS散度的取值范围是0～1，JS散度值越小则分布差异越小，匹配度分数越高，示例性地，计算公式可以是：匹配度分数＝1-min{JS散度}。In this embodiment, the target object in the image can be identified by the visual model, and the label corresponding to the target object can be obtained, which can be regarded as the keyword information of the image. In addition, the corresponding second category information and third category information can be obtained by querying the keyword knowledge base, and the JS divergence (Jensen-Shannon Divergence) is used to characterize the distribution difference between the two groups of category information, and the matching degree between the image content and the text content is calculated based on the minimum value of the distribution difference between the two. The value range of JS divergence is 0 to 1. The smaller the JS divergence value, the smaller the distribution difference and the higher the matching score. For example, the calculation formula can be: matching score = 1-min{JS divergence}.
更进一步地，请参考图5，步骤S150可以通过以下方法实现。Furthermore, referring to FIG. 5 , step S150 may be implemented by the following method.
步骤S151：对文字内容评分、图片内容评分及内容匹配度评分进行归一化处理，将文字内容评分、图片内容评分及内容匹配度评分映射到预设的分数范围。Step S151: normalizing the text content score, the picture content score and the content matching score, and mapping the text content score, the picture content score and the content matching score to a preset score range.
在本步骤中，为了消除不同分数指标之间的量纲影响，可以对数据进行标准化处理使其处于同一数值范围，具体地，可以先对内容评分、主题评分、情绪评分及通顺度评分进行归一化处理，得到文字内容评分，最后对文字内容评分、图片内容评分及内容匹配度评分进行归一化处理。In this step, in order to eliminate the dimensional influence between different score indicators, the data can be standardized so that it is in the same numerical range. Specifically, the content score, theme score, sentiment score and fluency score can be normalized first to obtain the text content score, and finally the text content score, image content score and content matching score can be normalized.
步骤S152：基于归一化处理后的文字内容评分、图片内容评分及内容匹配度评分得到一加权分数。Step S152: Obtain a weighted score based on the normalized text content score, image content score, and content matching score.
步骤S153：检测图文内容是否存在广告信息，当图文内容不存在广告信息时，将加权分数作为最终的评级分数，当图文内容存在广告信息时，对加权分数进行降分处理，将降分处理后的加权分数作为最终的评级分数。Step S153: Detect whether there is advertising information in the graphic content. When there is no advertising information in the graphic content, use the weighted score as the final rating score. When there is advertising information in the graphic content, downgrade the weighted score and use the weighted score after downgrading as the final rating score.
步骤S154：基于评级分数对图文内容进行评级，得到图文内容的评级结果。Step S154: Rating the graphic content based on the rating score to obtain a rating result of the graphic content.
在本实施例中，为了营造良好的图文社区，需要对广告贴进行一定的打压，当识别到图文内容存在广告信息时，可以对加权分数进行降分处理，以降低广告贴的权重，营造良好的社区氛围。基于评级分数可以综合多个维度全面地评估图文质量，对图文内容进行评级，提高评级结果的可信度和准确度。In this embodiment, in order to create a good picture and text community, it is necessary to suppress advertisements to a certain extent. When advertising information is identified in the picture and text content, the weighted score can be reduced to reduce the weight of the advertisement and create a good community atmosphere. Based on the rating score, the quality of the picture and text can be comprehensively evaluated from multiple dimensions, and the picture and text content can be rated to improve the credibility and accuracy of the rating results.
更进一步地，请参考图6，在步骤S150之后，本申请实施例提供的图文评级方法还可以包括以下方式步骤。Furthermore, please refer to FIG. 6 . After step S150 , the image and text rating method provided in the embodiment of the present application may further include the following steps.
步骤S161：基于图文内容的评级结果得到图文内容对应的推荐指数，并基于推荐指数对图文内容进行推荐。Step S161: obtaining a recommendation index corresponding to the graphic content based on the rating result of the graphic content, and recommending the graphic content based on the recommendation index.
步骤S162：基于图文内容的评级结果及评级分数生成对应的评级报告，并将评级报告发送至客户端。Step S162: Generate a corresponding rating report based on the rating results and rating scores of the graphic content, and send the rating report to the client.
在本实施例中，基于图文内容的评级结果对其赋予不同的推荐指数，可以对图文内容进行推荐排序，有助于筛选优质的图文内容，打压低质的图文内容，从而促进平台内容的良性循环，构建高质量的图文社区生态。此外，本方案还可以协助运营人员管理社区活动，提升运营效率。In this embodiment, different recommendation indexes are assigned to the graphic content based on its rating results, and the graphic content can be recommended and sorted, which helps to screen high-quality graphic content and suppress low-quality graphic content, thereby promoting a virtuous cycle of platform content and building a high-quality graphic community ecology. In addition, this solution can also assist operators in managing community activities and improve operational efficiency.
基于相同的发明构思，本申请实施还提供一种图文评级装置，请参照图7，图7为本申请实施例提供的图文评级装置的一种功能模块示意图。本申请实施例可以根据计算机设备执行的方法实施例对图文评级装置200进行功能模块的划分，也即该图文评级装置200所对应的以下各个功能模块可以用于执行上述各个方法实施例。其中，该图文评级装置200可以包括获取模块210、第一评分模块220、第二评分模块230、第三评分模块240以及评级模块250，下面分别对该图文评级装置200的各个功能模块的功能进行详细阐述。Based on the same inventive concept, the present application also provides a picture and text rating device. Please refer to Figure 7, which is a functional module schematic diagram of the picture and text rating device provided in the embodiment of the present application. The embodiment of the present application can divide the functional modules of the picture and text rating device 200 according to the method embodiment executed by the computer device, that is, the following functional modules corresponding to the picture and text rating device 200 can be used to execute the above-mentioned method embodiments. Among them, the picture and text rating device 200 can include an acquisition module 210, a first scoring module 220, a second scoring module 230, a third scoring module 240 and a rating module 250. The functions of each functional module of the picture and text rating device 200 are described in detail below.
获取模块210，用于获取待评级的图文内容，图文内容包括图片内容和文字内容。The acquisition module 210 is used to acquire the graphic content to be rated, where the graphic content includes picture content and text content.
待评级的图文内容包括，但不限于新闻、资讯及报告等，文字内容可以包括标题文本、正文文本等。The graphic and textual content to be rated includes, but is not limited to, news, information and reports, etc. The textual content may include title text, body text, etc.
本实施例中，获取模块210可以用于执行上述的步骤S110，关于获取模块210的详细实现方式可以参照上述针对步骤S110的详细描述。In this embodiment, the acquisition module 210 may be used to execute the above step S110. The detailed implementation of the acquisition module 210 may refer to the above detailed description of step S110.
第一评分模块220，用于检测文字内容中的关键词信息，并基于关键词信息对文字内容进行评分，得到文字内容评分。The first scoring module 220 is used to detect keyword information in the text content and score the text content based on the keyword information to obtain a text content score.
通过关键词信息可以从文字的表达、主题及情绪等维度对文字内容进行评估，从而提高评级结果的可靠性。Keyword information can be used to evaluate text content from dimensions such as text expression, theme, and emotion, thereby improving the reliability of rating results.
本实施例中，第一评分模块220可以用于执行上述的步骤S120，关于第一评分模块220的详细实现方式可以参照上述针对步骤S120的详细描述。In this embodiment, the first scoring module 220 may be used to execute the above step S120. The detailed implementation of the first scoring module 220 may refer to the above detailed description of step S120.
第二评分模块230，用于基于图片内容的图像质量对图片内容进行评分，得到图片内容评分。The second scoring module 230 is used to score the picture content based on the image quality of the picture content to obtain a picture content score.
由于图片内容的质量会影响图文内容的整体质量，故而可以基于美学质量和清晰度等维度对图片内容进行质量评估，示例性地，可以利用AVA美学评估数据集及KonIQ-10K数据集对深度学习模型训练得到的模型对图片内容进行评分，最终的图片内容评分可以是不同维度的评价分数的加权分数。Since the quality of image content will affect the overall quality of graphic content, the quality of image content can be evaluated based on dimensions such as aesthetic quality and clarity. For example, the model trained by the deep learning model using the AVA aesthetic evaluation dataset and the KonIQ-10K dataset can be used to score the image content. The final image content score can be a weighted score of the evaluation scores of different dimensions.
本实施例中，第二评分模块230可以用于执行上述的步骤S130，关于第二评分模块230的详细实现方式可以参照上述针对步骤S130的详细描述。In this embodiment, the second scoring module 230 may be used to execute the above step S130. The detailed implementation of the second scoring module 230 may refer to the above detailed description of step S130.
第三评分模块240，用于基于图片内容与文字内容的匹配程度进行评分，得到内容匹配度评分。The third scoring module 240 is used to score based on the matching degree between the image content and the text content to obtain a content matching score.
通过考量文字内容与图片内容的匹配程度，可以全面评估图文内容的一致性和协调性，从而更为全面、准确地评估整体的图文质量，提高评级结果的可信度。By considering the degree of match between text content and image content, we can comprehensively evaluate the consistency and coordination of the text and image content, thereby more comprehensively and accurately evaluating the overall quality of the image and text, and improving the credibility of the rating results.
本实施例中，第三评分模块240可以用于执行上述的步骤S140，关于第三评分模块240的详细实现方式可以参照上述针对步骤S140的详细描述。In this embodiment, the third scoring module 240 may be used to execute the above step S140. The detailed implementation of the third scoring module 240 may refer to the above detailed description of step S140.
评级模块250，用于基于文字内容评分、图片内容评分及内容匹配度评分对图文内容进行评级。The rating module 250 is used to rate the graphic content based on the text content rating, the image content rating and the content matching rating.
基于文字内容评分、图片内容评分和内容匹配度评分对图文内容进行评级，可以综合不同模态的信息全面、综合地评估图文质量，使得评级结果更具可信度及准确度。By rating the text content based on the text content rating, image content rating and content matching rating, we can comprehensively and integratedly evaluate the quality of the text and image by integrating information from different modalities, making the rating results more credible and accurate.
本实施例中，评级模块250可以用于执行上述的步骤S150，关于评级模块250的详细实现方式可以参照上述针对步骤S150的详细描述。In this embodiment, the rating module 250 can be used to execute the above step S150. The detailed implementation of the rating module 250 can refer to the detailed description of the above step S150.
在本实施例中，通过多模块对文字内容、图片内容和图文内容的匹配度等进行评分，并基于文字内容评分、图片内容评分和内容匹配度评分对图文内容进行评级，从而可以综合不同模态的信息全面、综合地评估图文质量，提高评级结果的可信度及准确度。此外，本方案无需多模态标签数据和高质量图像-文本数据对训练模型，且在实际预测过程中并不约束文本中的关键词一定出现于图像中，不仅可以节约人力资源，还可以提高模型的性能，提高评级结果的可信度及准确度。In this embodiment, the text content, image content and the matching degree of image and text content are scored through multiple modules, and the image and text content is rated based on the text content score, image content score and content matching score, so that the image and text quality can be comprehensively and comprehensively evaluated by integrating information from different modalities, and the credibility and accuracy of the rating results can be improved. In addition, this solution does not require multi-modal label data and high-quality image-text data to train the model, and in the actual prediction process, it does not constrain the keywords in the text to appear in the image, which can not only save human resources, but also improve the performance of the model and improve the credibility and accuracy of the rating results.
需要说明的是，应理解以上装置或系统中的各个模块的划分仅仅是一种逻辑功能的划分，实际实现时可以全部或部分集成到一个物理实体上，也可以在物理上分开。且这些模块可以全部以软件(比如，开源软件)可以通过处理器调用的形式实现；也可以全部以硬件的形式实现；还可以部分模块通过处理器调用软件的形式实现，部分模块通过硬件的形式实现。作为一种示例，第一评分模块220可以由单独处理器运行实现，可以以程序代码的形式存储于上述装置或系统的存储器中，由上述装置或系统的某一个处理器调用并执行以上第一评分模块220的功能，其它模块的实现与之类似，在此就不再赘述。此外这些模块可以全部或部分集成在一起，也可以独立实现。这里所描述的处理器可以是一种具有信号的处理能力的集成电路，在实现过程中，上述技术方案中的各步骤或各个模块可以通过处理器中的集成逻辑电路或者执行软件程序的形式完成。It should be noted that it should be understood that the division of the various modules in the above device or system is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. And these modules can all be implemented in the form of software (for example, open source software) that can be called by the processor; they can also be implemented in the form of hardware; some modules can also be implemented in the form of software called by the processor, and some modules can be implemented in the form of hardware. As an example, the first scoring module 220 can be implemented by running a separate processor, and can be stored in the memory of the above device or system in the form of program code. The function of the above first scoring module 220 is called and executed by a processor of the above device or system. The implementation of other modules is similar to it, which will not be repeated here. In addition, these modules can be fully or partially integrated together, or they can be implemented independently. The processor described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step or each module in the above technical solution can be completed by an integrated logic circuit in the processor or by executing a software program.
请参照图8，图8示出了本公开实施例提供的用于实现上述的图文评级方法的服务器300的硬件结构示意图。如图8所示，服务器300可包括处理器310、计算机可读存储介质320、总线330及通信单元340。Please refer to Fig. 8, which shows a hardware structure diagram of a server 300 for implementing the above-mentioned image and text rating method provided by an embodiment of the present disclosure. As shown in Fig. 8, the server 300 may include a processor 310, a computer-readable storage medium 320, a bus 330 and a communication unit 340.
在具体实现过程中，处理器310执行计算机可读存储介质320存储的计算机执行指令(例如图7中所示的图文评级装置200中的各个模块)，使得处理器310可以执行如上方法实施例的图文评级方法，其中，处理器310、计算机可读存储介质320以及通信单元340可以通过总线330连接。During the specific implementation process, the processor 310 executes computer-executable instructions stored in the computer-readable storage medium 320 (for example, the various modules in the image and text rating device 200 shown in Figure 7), so that the processor 310 can execute the image and text rating method of the above method embodiment, wherein the processor 310, the computer-readable storage medium 320 and the communication unit 340 can be connected via the bus 330.
处理器310的具体实现过程可参见上述服务器300执行的各个方法实施例，其实现原理和技术效果类似，本申请实施例此处不再赘述。The specific implementation process of the processor 310 can refer to the various method embodiments executed by the above-mentioned server 300. The implementation principles and technical effects are similar, and the embodiments of the present application will not be repeated here.
计算机可读存储介质320可以是，但不限于，随机存取存储器(Random AccessMemory，RAM)，只读存储器(Read Only Memory，ROM)，可编程只读存储器(ProgrammableRead-Only Memory，PROM)，可擦除只读存储器(Erasable Programmable Read-OnlyMemory，EPROM)，电可擦除只读存储器(Electric Erasable Programmable Read-OnlyMemory，EEPROM)等。其中，存储器111用于存储程序或者数据。The computer readable storage medium 320 may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), etc. Among them, the memory 111 is used to store programs or data.
总线330可以分为地址总线、数据总线、控制总线等。为便于表示，本申请附图中的总线并不限定仅有一根总线或一种类型的总线。The bus 330 can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, the buses in the drawings of the present application are not limited to only one bus or one type of bus.
在本申请实施例提供的交互场景中，通信单元340可用于与客户端通信，以实现服务器300与客户端之间的数据交互。In the interaction scenario provided in the embodiment of the present application, the communication unit 340 can be used to communicate with the client to implement data interaction between the server 300 and the client.
此外，本申请实施例还提供一种可读存储介质，可读存储介质中存储有计算机执行指令，当处理器执行计算机执行指令时，实现如上的图文评级方法。In addition, an embodiment of the present application further provides a readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above-mentioned image and text rating method is implemented.
综上所述，本申请实施例提供一种图文评级方法、装置及计算机可读存储，通过对文字内容、图片内容和图文内容的匹配度等进行评分，并基于文字内容评分、图片内容评分和内容匹配度评分对图文内容进行评级，从而可以综合不同模态的信息全面、综合地评估图文质量，提高评级结果的可信度及准确度。此外，本方案无需多模态标签数据和高质量图像-文本数据对训练模型，且在实际预测过程中并不约束文本中的关键词一定出现于图像中，不仅可以节约人力资源，还可以提高模型的性能，提高评级结果的可信度及准确度。In summary, the embodiments of the present application provide a method, device and computer-readable storage for rating text and images, which rates text and image content based on text content rating, image content rating and content matching rating, thereby comprehensively and comprehensively evaluating the quality of text and images by integrating information from different modalities, and improving the credibility and accuracy of rating results. In addition, this solution does not require multimodal label data and high-quality image-text data to train the model, and in the actual prediction process, it does not constrain the keywords in the text to appear in the image, which can not only save human resources, but also improve the performance of the model and the credibility and accuracy of the rating results.
以上所述仅为本申请的优选实施例而已，并不用于限制本申请，对于本领域的技术人员来说，本申请可以有各种更改和变化。凡在本申请的精神和原则之内，所作的任何修改、等同替换、改进等，均应包含在本申请的保护范围之内。The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410011782.2ACN118094239A (en) | 2024-01-03 | 2024-01-03 | Image and text rating method, device and computer-readable storage medium |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410011782.2ACN118094239A (en) | 2024-01-03 | 2024-01-03 | Image and text rating method, device and computer-readable storage medium |
| Publication Number | Publication Date |
|---|---|
| CN118094239Atrue CN118094239A (en) | 2024-05-28 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410011782.2APendingCN118094239A (en) | 2024-01-03 | 2024-01-03 | Image and text rating method, device and computer-readable storage medium |
| Country | Link |
|---|---|
| CN (1) | CN118094239A (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119204852A (en)* | 2024-11-28 | 2024-12-27 | 宁波深擎信息科技有限公司 | A method, device and equipment for evaluating the quality of information content in the financial and economic field |
| CN119250655A (en)* | 2024-12-05 | 2025-01-03 | 粤港澳大湾区数字经济研究院(福田) | Object rating method, device, equipment and storage medium |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543512A (en)* | 2018-10-09 | 2019-03-29 | 中国科学院自动化研究所 | The evaluation method of picture and text abstract |
| CN109800349A (en)* | 2018-12-17 | 2019-05-24 | 北京邮电大学 | The data processing method and device of content quantization news value are issued based on user |
| CN110119473A (en)* | 2019-05-23 | 2019-08-13 | 北京金山数字娱乐科技有限公司 | A kind of construction method and device of file destination knowledge mapping |
| CN111104789A (en)* | 2019-11-22 | 2020-05-05 | 华中师范大学 | Text scoring method, device and system |
| CN111882371A (en)* | 2019-04-15 | 2020-11-03 | 阿里巴巴集团控股有限公司 | Content information processing method, image-text content processing method, computer device, and medium |
| CN112435064A (en)* | 2020-11-27 | 2021-03-02 | 北京沃东天骏信息技术有限公司 | Method, device and equipment for evaluating recommendation information and computer readable storage medium |
| CN114818691A (en)* | 2021-01-29 | 2022-07-29 | 腾讯科技（深圳）有限公司 | Evaluation method, device, equipment and medium of article content |
| CN116385856A (en)* | 2023-03-22 | 2023-07-04 | 鹏城实验室 | Data transmission method, device and storage medium |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543512A (en)* | 2018-10-09 | 2019-03-29 | 中国科学院自动化研究所 | The evaluation method of picture and text abstract |
| CN109800349A (en)* | 2018-12-17 | 2019-05-24 | 北京邮电大学 | The data processing method and device of content quantization news value are issued based on user |
| CN111882371A (en)* | 2019-04-15 | 2020-11-03 | 阿里巴巴集团控股有限公司 | Content information processing method, image-text content processing method, computer device, and medium |
| CN110119473A (en)* | 2019-05-23 | 2019-08-13 | 北京金山数字娱乐科技有限公司 | A kind of construction method and device of file destination knowledge mapping |
| CN111104789A (en)* | 2019-11-22 | 2020-05-05 | 华中师范大学 | Text scoring method, device and system |
| CN112435064A (en)* | 2020-11-27 | 2021-03-02 | 北京沃东天骏信息技术有限公司 | Method, device and equipment for evaluating recommendation information and computer readable storage medium |
| CN114818691A (en)* | 2021-01-29 | 2022-07-29 | 腾讯科技（深圳）有限公司 | Evaluation method, device, equipment and medium of article content |
| CN116385856A (en)* | 2023-03-22 | 2023-07-04 | 鹏城实验室 | Data transmission method, device and storage medium |
| Title |
|---|
| 戴志强主编: "《艺术与科学研究 2012.2》", 31 December 2012, 北京：中国广播电视出版社, pages: 92 - 93* |
| 通证一哥: "《你好ChatGPT》", 30 April 2023, 北京：机械工业出版社, pages: 78* |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119204852A (en)* | 2024-11-28 | 2024-12-27 | 宁波深擎信息科技有限公司 | A method, device and equipment for evaluating the quality of information content in the financial and economic field |
| CN119250655A (en)* | 2024-12-05 | 2025-01-03 | 粤港澳大湾区数字经济研究院(福田) | Object rating method, device, equipment and storage medium |
| Publication | Publication Date | Title |
|---|---|---|
| US10567329B2 (en) | Methods and apparatus for inserting content into conversations in on-line and digital environments | |
| Gu et al. | " what parts of your apps are loved by users?"(T) | |
| Batool et al. | Precise tweet classification and sentiment analysis | |
| CN102163187B (en) | Document marking method and device | |
| CN111930792B (en) | Labeling method and device for data resources, storage medium and electronic equipment | |
| KR20160055930A (en) | Systems and methods for actively composing content for use in continuous social communication | |
| CN118094239A (en) | Image and text rating method, device and computer-readable storage medium | |
| CN112699645B (en) | Corpus labeling method, apparatus and device | |
| CN107657056A (en) | Method and apparatus based on artificial intelligence displaying comment information | |
| Ferschke et al. | A survey of nlp methods and resources for analyzing the collaborative writing process in wikipedia | |
| CN110162597A (en) | Article data processing method, device, computer-readable medium and electronic equipment | |
| CN105956181A (en) | Searching method and apparatus | |
| CN104881447A (en) | Searching method and device | |
| Joseph et al. | Exploring the application of natural language processing for social media sentiment analysis | |
| CN113268651B (en) | A method and device for automatically generating a summary of search information | |
| CN114168837A (en) | Chatbot searching method, equipment and storage medium | |
| CN108717450A (en) | Film review emotional orientation analysis algorithm | |
| CN110705257B (en) | Media resource identification method and device, storage medium and electronic device | |
| CN116955697A (en) | Analysis method, analysis device, analysis equipment, analysis medium and analysis program product for search results | |
| CN116153496A (en) | Neural network model training method and depression emotion detection method | |
| Dhanalakshmi et al. | Automated Sentiment Analysis for Instant Feedback on YouTube videos through comments | |
| CN108763203B (en) | A method of using feature word set to represent movie reviews as feature vectors in film review sentiment analysis | |
| CN115130453A (en) | Interactive information generation method and device | |
| JP7703599B2 (en) | Opinion analysis system, opinion analysis method, and program | |
| Le et al. | Applying Artificial Neural Network for Sentiment Analytics of Social Media Text Data in fastfood industry |
| 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 |