Content quality assessment method and system

Abstract

The invention relates to a content quality evaluation method and system. The method comprises the steps of obtaining to-be-evaluated content, and calling a pre-trained evaluation model; the evaluation model is obtained by training a plurality of historical published contents; inputting the to-be-evaluated content into an evaluation model, and calculating a content quality score of the to-be-evaluated content through the evaluation model; and based on the content quality score, evaluating the quality level of the to-be-evaluated content. According to the scheme provided by the invention, the content quality evaluation efficiency and accuracy can be improved, the labor cost is reduced, and the scale is expanded.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to content quality assessment methods and systems. Background Technology

[0002] Content quality assessment can effectively mitigate risks, enhance value, and ensure the effective use of resources; therefore, it is an indispensable step.

[0003] In related technologies, content quality assessment mainly relies on manual review. However, this method is inefficient, costly, and difficult to scale. Furthermore, it is highly dependent on the reviewer's experience and is prone to inconsistent assessment standards due to subjective human factors, resulting in uneven content quality. Summary of the Invention

[0004] To address or partially address the problems existing in related technologies, this application provides a content quality assessment method and system that can improve the efficiency and accuracy of content quality assessment, reduce labor costs, and expand scale.

[0005] The first aspect of this application provides a content quality assessment method, including: The system retrieves the content to be evaluated and calls a pre-trained evaluation model; the evaluation model is trained using multiple historical published content. The content to be evaluated is input into the evaluation model to calculate the content quality score of the content to be evaluated. The quality level of the content to be evaluated is determined based on the content quality score.

[0006] In one embodiment, calculating the content quality score of the content to be evaluated using the evaluation model includes: The quality index score of the content to be evaluated is calculated using the evaluation model; the quality index score includes at least one of the following: uniqueness score, topicality score, readability score, completeness score, authenticity score, and brand friendliness score. Determine the content type and brand strategy of the content to be evaluated; Based on the content type and the brand strategy, assign corresponding weights to the scores of each of the quality indicators; The weights corresponding to the scores of each of the aforementioned quality indicators are input into the evaluation model, so that the evaluation model can calculate the content quality score of the content to be evaluated using the uniqueness score, the topicality score, the readability score, the completeness score, the authenticity score, the brand friendliness score, and the corresponding weights.

[0007] In one embodiment, assessing the quality level of the content to be evaluated based on the content quality score includes: If the content quality score is greater than or equal to the first threshold, and all the scores of the quality indicators are greater than or equal to the second threshold, then the quality level of the content to be evaluated is assessed as the first level; or, If the content quality score is greater than or equal to the first threshold, and there are N quality indicator scores that are all less than the second threshold, then the quality level of the content to be evaluated is assessed as the second level, where N is a positive integer less than 6; or, If the content quality score is less than the first threshold, the quality level of the content to be evaluated will be assessed as the third level. Among them, the first level > the second level > the third level.

[0008] In one embodiment, calculating the quality index score of the content to be evaluated using the evaluation model includes: The similarity between the content to be evaluated and each of the historical published contents is calculated using the evaluation model. The evaluation model is used to sort the multiple similarities in descending order to select the target similarity that is first. The evaluation model calculates the uniqueness score of the content to be evaluated based on the target similarity; wherein the uniqueness score is negatively correlated with the target similarity.

[0009] In one embodiment, calculating the quality index score of the content to be evaluated using the evaluation model includes: The evaluation model is used to extract the popularity characteristics, timeliness characteristics, and interaction potential characteristics of the content to be evaluated. The evaluation model calculates the topicality score of the content to be evaluated based on the popularity feature, the timeliness feature, and the interaction potential feature; wherein the topicality score is positively correlated with the popularity feature, the timeliness feature, and the interaction potential feature.

[0010] In one embodiment, calculating the quality index score of the content to be evaluated using the evaluation model includes: The evaluation model determines the readability of the content to be evaluated based on the average sentence length and average word length; and, The narrative structure completeness of the content to be evaluated is determined by the evaluation model based on the chapter markers of the content to be evaluated. The evaluation model calculates the readability score of the content to be evaluated based on the reading friendliness and the narrative structure completeness; wherein the readability score is positively correlated with the reading friendliness and the narrative structure completeness.

[0011] In one embodiment, calculating the quality index score of the content to be evaluated using the evaluation model includes: The evaluation model is used to calculate the element completeness of the content to be evaluated, which contains preset key elements; wherein, different content types correspond to different preset key elements. The evaluation model calculates the completeness score of the content to be evaluated based on the element completeness; the completeness score is positively correlated with the element completeness.

[0012] In one embodiment, calculating the quality index score of the content to be evaluated using the evaluation model includes: The factual statements in the content to be evaluated are extracted using the evaluation model. The factual statements are verified using the official database through the evaluation model, and the verification results are obtained. The evaluation model calculates the authenticity score of the content to be evaluated based on the verification results.

[0013] In one embodiment, calculating the quality index score of the content to be evaluated using the evaluation model includes: Sentiment analysis is performed on the content to be evaluated using the evaluation model to obtain a positive sentiment score for the content to be evaluated; and, The risk word density of the content to be evaluated containing preset risk words is calculated using the evaluation model. The evaluation model calculates the brand friendliness score of the content to be evaluated based on the positive sentiment score and the risk word density; wherein the brand friendliness is positively correlated with the positive sentiment score and negatively correlated with the risk word density.

[0014] A second aspect of this application provides a content quality assessment system, comprising: The acquisition module is used to acquire the content to be evaluated and call a pre-trained evaluation model; the evaluation model is trained using multiple historical published contents. The calculation module is used to input the content to be evaluated into the evaluation model, so as to calculate the content quality score of the content to be evaluated through the evaluation model; The evaluation module is used to evaluate the quality level of the content to be evaluated based on the content quality score.

[0015] A third aspect of this application provides an electronic device, comprising: Processor; and A memory that stores executable code, which, when executed by the processor, causes the processor to perform the method described above.

[0016] A fourth aspect of this application provides a computer-readable storage medium having executable code stored thereon, which, when executed by a processor of an electronic device, causes the processor to perform the method described above.

[0017] The fifth aspect of this application provides a computer program product comprising computer instructions that, when executed by a processor, implement the method described above.

[0018] The technical solution provided in this application may include the following beneficial results: The solution provided in this application obtains the content to be evaluated and calls a pre-trained evaluation model. The evaluation model is trained using multiple historically published content. The content to be evaluated is input into the evaluation model to calculate a content quality score. Based on the content quality score, the quality level of the content to be evaluated is assessed. This application calculates the content quality score of the content to be evaluated through the evaluation model. The content quality score can effectively measure the value of the content itself. Therefore, based on the content quality score, the quality level of the content to be evaluated can be accurately assessed. The entire process is automated, with millisecond-level response, and is entirely driven by data and algorithms, thereby improving the efficiency and accuracy of content quality evaluation, reducing labor costs, and expanding scalability.

[0019] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0020] The above and other objects, features and advantages of this application will become more apparent from the more detailed description of exemplary embodiments thereof in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments thereof.

[0021] Figure 1 This is a schematic flowchart illustrating the content quality assessment method in an embodiment of this application; Figure 2 This is another schematic diagram of the content quality assessment method shown in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the content quality assessment system shown in the embodiments of this application; Figure 4 This is a schematic diagram of the structure of an electronic device shown in an embodiment of this application. Detailed Implementation

[0022] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to make this application more thorough and complete, and to fully convey the scope of this application to those skilled in the art.

[0023] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0024] It should be understood that although the terms "first," "second," "third," etc., may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0025] In related technologies, content quality assessment mainly relies on manual review. For example, reviewers assess content quality based on social media display data (such as the number of likes, reposts, comments, and favorites). However, this method is inefficient, costly, and difficult to scale. Moreover, this method is highly dependent on the reviewer's experience and is prone to inconsistent assessment standards due to human subjective factors, resulting in uneven content quality.

[0026] To address the aforementioned issues, this application provides a content quality assessment method. The method calculates a content quality score for the content to be assessed using an assessment model. This score effectively measures the value of the content itself, thus accurately determining its quality level. The entire process is automated, with millisecond-level response, and is entirely driven by data and algorithms, thereby improving the efficiency and accuracy of content quality assessment, reducing labor costs, and expanding its scale.

[0027] The technical solutions of the embodiments of this application are described in detail below with reference to the accompanying drawings.

[0028] Figure 1This is a flowchart illustrating the content quality assessment method in an embodiment of this application.

[0029] See Figure 1 The content quality assessment method of this application may include: S110: Obtain the content to be evaluated and call the pre-trained evaluation model; the evaluation model is trained using multiple historical published contents.

[0030] In this application embodiment, the content quality assessment system (hereinafter referred to as the "assessment system") can be applied. Before content is published to various social media platforms, the assessment system provided in this application embodiment can be used to assess the quality of the content. For ease of distinction, the content can be defined as the content to be assessed. In specific implementation, the assessment system can crawl or receive the content to be assessed from various social media platforms through an API (Application Programming Interface). The content format of the content to be assessed may include at least one of text, images, graphic text, video, and audio.

[0031] This application embodiment can crawl or receive multiple historical posts from various social media platforms via API interfaces. The historical posts and the content to be evaluated have the same content format; the difference is that the content to be evaluated is content not published on social media platforms, while the historical posts are content already published on social media platforms. This application embodiment can pre-train the evaluation model using multiple historical posts, and then deploy the trained evaluation model to the evaluation system so that it can be directly invoked to evaluate the content in practical applications. The evaluation models include TF-IDF (Term Frequency - Inverse Document Frequency), BERT (Bidirectional Encoder Representations from Transformers), LSTM (Long Short-Term Memory), dependency parsing, sequence labeling models, NLP (Natural Language Processing), and SKEP (Sentiment Knowledge Enhanced Pre-training).

[0032] S120, Input the content to be evaluated into the evaluation model so that the content quality score of the content to be evaluated can be calculated by the evaluation model.

[0033] The evaluation system can input the content to be evaluated into the evaluation model so that the evaluation model can extract and calculate the features of multi-dimensional indicators of the content to be evaluated, and then calculate the final comprehensive score based on the multi-dimensional indicators. The final comprehensive score can be called the content quality score, which can be used to measure the value of the content to be evaluated itself.

[0034] S130 assesses the quality level of the content to be evaluated based on the content quality score.

[0035] The content quality score calculated by the evaluation model is intended to measure the value of the content itself. It is closely related to the brand, industry, product, and other characteristics of the content. It is more representative and authoritative than the results measured by social media display data (such as likes, reposts, comments, and collections) in related technologies. Therefore, the evaluation system can accurately assess the quality level of the content based on the content quality score. The higher the content quality score, the higher the value of the content itself, and the higher the quality level of the content.

[0036] It should be noted that the application scenarios of this application embodiment may include AIUC (AI Understand Content). AIUC is an AI understanding system that focuses on building the ability to understand consumer content, and can read, evaluate and restore the value of content.

[0037] As this example illustrates, the solution provided in this application involves acquiring the content to be evaluated and calling a pre-trained evaluation model. The evaluation model is trained using multiple historically published content samples. The content to be evaluated is input into the evaluation model to calculate its content quality score. Based on this score, the quality level of the content is assessed. This application calculates the content quality score using an evaluation model. The content quality score effectively measures the value of the content itself, thus accurately assessing its quality level. The entire process is automated, with millisecond-level response, and is entirely data- and algorithm-driven, thereby improving the efficiency and accuracy of content quality assessment, reducing labor costs, and scalability.

[0038] Figure 2 This is another flowchart illustrating the content quality assessment method shown in this application.

[0039] See Figure 2 The content quality assessment method of this application may include: S210: Obtain the content to be evaluated and call the pre-trained evaluation model; the evaluation model is trained using multiple historical published contents.

[0040] This step can be found in the description in S110, and will not be repeated here.

[0041] S220, Input the content to be evaluated into the evaluation model to calculate the quality index score of the content to be evaluated through the evaluation model; the quality index score includes at least one of the following: uniqueness score, topicality score, readability score, completeness score, authenticity score, and brand friendliness score.

[0042] In addition, some related technologies rely on automated review methods, such as automatically filtering out content containing sensitive words or typos based on keywords. However, this method is essentially a passive "gatekeeper" filter that cannot accurately understand the semantics, context, and deeper value of the content. Automated evaluation based on a single dimension (such as keyword filtering) cannot comprehensively measure the value of the content.

[0043] To address this, this application's embodiments cover six core dimensions with multi-dimensional indicators: uniqueness, topicality, readability, completeness, authenticity, and brand friendliness, thereby employing these multi-dimensional indicators to comprehensively measure content value. In specific implementation, the evaluation system can input the content to be evaluated into the evaluation model, allowing the model to extract and calculate features of the multi-dimensional indicators. These multi-dimensional indicators can be represented by quality indicator scores, which may include at least one of the following: uniqueness score, topicality score, readability score, completeness score, authenticity score, and brand friendliness score.

[0044] In one embodiment, calculating the quality index score of the content to be evaluated using an evaluation model may include: The evaluation model calculates the similarity between the content to be evaluated and each historical post. The evaluation model sorts the multiple similarities in descending order to select the top target similarity. Based on the target similarity, the evaluation model calculates the uniqueness score of the content to be evaluated. The uniqueness score is negatively correlated with the target similarity.

[0045] Regarding the core dimension of uniqueness, the evaluation criteria include originality, information increment, and scarcity. This application embodiment can employ NLP technology to calculate text similarity, thereby calculating the uniqueness score of the content to be evaluated based on text similarity. In specific implementation, if the content to be evaluated is in a non-text format (such as images, text-image pairs, videos, or audio), the evaluation system can first perform OCR (Optical Character Recognition) on the content to extract the text portion. Then, the text to be evaluated (i.e., the text-based content) is preprocessed, including at least one of word segmentation and stop word removal. The preprocessed text is then input into the evaluation model (specifically, a pre-trained model such as TF-IDF or BERT) to vectorize the preprocessed text using TF-IDF or BERT, i.e., converting the text into vectors. Then, tools such as Faiss (similarity search) and Elasticsearch (keyword search) can be used to perform vector similarity retrieval in a pre-defined content library. Specifically, the pre-defined content library contains multiple historical published content items. Similarly, if any historical published content is in a non-text format (such as images, text-image, video, or audio), the evaluation system can first perform OCR on that historical published content to extract the text portion. Then, each historical published text (i.e., historical published content in text form) is preprocessed. TF-IDF or BERT is used to vectorize the features of each preprocessed historical published text, converting each historical published text into a vector. TF-IDF or BERT is then used to calculate the similarity between the vector of the text to be evaluated and the vector of each historical published text. TF-IDF or BERT can sort these similarities in descending order to select the highest target similarity. The target similarity is then the highest similarity. Based on the target similarity, TF-IDF or BERT calculates the uniqueness score of the content to be evaluated. The specific formula is: Uniqueness Score = 1 - Target Similarity. It can be seen that the uniqueness score is negatively correlated with the target similarity; that is, the lower the target similarity, the higher the uniqueness score.

[0046] In one embodiment, calculating the quality index score of the content to be evaluated using an evaluation model may include: The evaluation model extracts the popularity, timeliness, and interaction potential characteristics of the content to be evaluated. Based on these characteristics, the evaluation model calculates the topicality score of the content to be evaluated. The topicality score is positively correlated with the popularity, timeliness, and interaction potential characteristics.

[0047] Regarding the core dimension of topicality, the evaluation criteria include popularity, timeliness, and interaction potential. This application embodiment can integrate internal data and external trend data to calculate the topicality score of the content to be evaluated based on these characteristic data. In its implementation, the evaluation model can extract the popularity features, timeliness features, and interaction potential features of the content to be evaluated. Specifically, for the extraction of popularity features, the specific operation can be to directly collect explicit social media data (such as likes, comments, and reposts) within a certain period after the content to be evaluated is published, and then normalize this explicit social media data to obtain the popularity features of the content to be evaluated. For the extraction of timeliness features, the specific operation can be to identify time entities (such as "today" or "2023") in the text to be evaluated, or to calculate the interval between the publication time and the current time; the shorter the interval, the higher the score, thus obtaining the timeliness features of the content to be evaluated. For the extraction of interaction potential features, the specific operation can be to use the evaluation model (specifically, LSTM or BERT) to predict the probability of the text to be evaluated triggering comments / reposts. It should be noted that, in this embodiment, the parameters of the evaluation model can be updated using recently collected historical published content (text -> subsequent interaction volume) to improve the prediction accuracy. After extracting the popularity characteristics, timeliness characteristics, and interaction potential characteristics of the content to be evaluated, the evaluation model can calculate the topicality score of the content to be evaluated based on these characteristics. The topicality score is positively correlated with the popularity characteristics, timeliness characteristics, and interaction potential characteristics. That is, the higher the score of the popularity characteristics, and / or the higher the score of the timeliness characteristics, and / or the higher the score of the interaction potential characteristics, the higher the topicality score.

[0048] In one embodiment, calculating the quality index score of the content to be evaluated using an evaluation model may include: The evaluation model determines the readability of the content to be evaluated based on the average sentence length and average word length; it also determines the narrative structure completeness of the content to be evaluated based on the text markers; and calculates the readability score of the content to be evaluated based on the readability and narrative structure completeness. The readability score is positively correlated with both readability and narrative structure completeness.

[0049] For the core dimension of readability, the evaluation criteria include narrative structural integrity and reading friendliness. This application's embodiments can utilize readability formulas and syntactic analysis to calculate the readability score of the content to be evaluated. In a specific implementation, the evaluation model can use the Flesch-Kincaid (text readability formula) to calculate the readability index of the text to be evaluated based on the average sentence length and average word length. This index determines the reading friendliness of the content to be evaluated. The readability index is positively correlated with reading friendliness; a higher readability index indicates that the content is easier to read, thus indicating higher reading friendliness. Simultaneously, the evaluation model (specifically, dependency parsing or sequence labeling models) identifies whether the text to be evaluated contains specific discourse markers such as introductory openings (e.g., "firstly," "notably"), logical connectors (e.g., "therefore," "however"), and summarizing endings. The more of these specific discourse markers a text contains, the higher its narrative structural integrity. After determining the readability and narrative structure completeness of the content to be evaluated, the evaluation model can calculate the readability score based on these two factors. The specific calculation formula is: Readability Score = Readability + Narrative Structure Completeness. It can be seen that the readability score is positively correlated with both readability and narrative structure completeness. That is, the higher the readability and / or the higher the narrative structure completeness, the higher the readability score.

[0050] In one embodiment, calculating the quality index score of the content to be evaluated using an evaluation model may include: The evaluation model calculates the element completeness of the content to be evaluated, which includes preset key elements. Different content types correspond to different preset key elements. Based on the element completeness, the evaluation model calculates the completeness score of the content to be evaluated. The completeness score is positively correlated with the element completeness.

[0051] For the core dimension of completeness, the evaluation criteria include information density and element completeness. This application embodiment can construct a checklist based on domain knowledge to calculate the completeness score of the content to be evaluated. In specific implementation, this application embodiment can define preset key elements (Who, What, When, Where, Why, How) that should be included according to different content types (such as product reviews, news, tutorials). For example, a product review (such as a mobile phone review) should include preset key elements such as "appearance (What)," "performance (How)," "photography (How)," "battery life (How / When)," and "price (Why)." A news article should include preset key elements such as "time (When)," "location (Where)," "people (Who)," "event (What)," and "reason (Why)." This allows for the use of Named Entities Responsible for (NER) in practical applications. Named entity recognition (NIR) technology and keyword matching are used to detect whether the text to be evaluated contains these preset key elements. The number of preset key elements detected is then counted, which is the element completeness. The completeness score of the content to be evaluated is obtained by taking the ratio of the number of preset key elements detected (i.e., element completeness) to the total number of preset key elements that should be included. The specific calculation formula is: Completeness score = Number of preset key elements detected / Total number of preset key elements that should be included. It can be seen that the completeness score is positively correlated with the number of preset key elements detected (i.e., element completeness). That is, the more preset key elements detected (or the higher the element completeness), the higher the completeness score.

[0052] In one embodiment, calculating the quality index score of the content to be evaluated using an evaluation model may include: The evaluation model extracts factual statements from the content to be evaluated; the evaluation model verifies the factual statements using an official database to obtain the verification results; and the evaluation model calculates the authenticity score of the content to be evaluated based on the verification results.

[0053] Regarding the core dimension of authenticity, the evaluation criteria include factual accuracy and logical consistency. This application's embodiments can combine knowledge graphs and trusted source verification to calculate the authenticity score of the content to be evaluated. In a specific implementation, the evaluation model can extract claimed factual statements from the text to be evaluated (e.g., "The battery capacity of a XX brand mobile phone is 5000mAh"), and then access official databases (such as Wikipedia, official brand websites, and authoritative media information) as trusted sources (or as knowledge bases) to verify the factual statements. Based on these verification results, the authenticity score of the content to be evaluated is calculated. Specifically, the evaluation model (which can be an NLP model) can compare the extracted factual statements with the trusted sources (i.e., the official database) to determine whether there are contradictions in the statements within the text to be evaluated. If no contradictions are found, the content is logically consistent, and the authenticity score is the first value. If contradictions are found, the content is logically inconsistent, and the authenticity score is the second value. The first value is greater than the second value.

[0054] In one embodiment, calculating the quality index score of the content to be evaluated using an evaluation model may include: The evaluation model performs sentiment analysis on the content to be evaluated to obtain a positive sentiment score; it also calculates the risk word density of the content to be evaluated containing preset risk words; and it calculates the brand friendliness score of the content to be evaluated based on the positive sentiment score and the risk word density. Among these, brand friendliness is positively correlated with the positive sentiment score, and brand friendliness is negatively correlated with the risk word density.

[0055] For the core dimension of brand friendliness, the assessment points include sentiment orientation and risk aversion. This application's embodiments can perform sentiment analysis and high-risk pattern recognition on the content to be evaluated to calculate its brand friendliness. In specific implementation, the assessment model (specifically, the SKEP model) determines the overall positive or negative sentiment of the text to be evaluated. Specifically, this can be done by identifying negative statements (such as "I hate XX") within the text, then counting the number of identified negative statements, and calculating a positive sentiment score based on the number of negative statements. The number of negative statements is negatively correlated with the positive sentiment score; that is, the more negative statements, the lower the positive sentiment score. Simultaneously, the assessment model can identify whether the text contains negative statements targeting a specific brand (such as "Some brands perform poorly in certain aspects"). The more negative statements targeting a specific brand the text contains, the lower the positive sentiment score. Furthermore, this application embodiment can pre-establish a risk terminology database, which includes preset risk terms such as abusive words, sensitive topics, and competitor keywords. The preset risk terms in the database can be updated periodically. The evaluation model can count the number of preset risk terms from the database contained in the text to be evaluated. Risk term density = number of preset risk terms contained in the text to be evaluated / total number of words contained in the text to be evaluated. After calculating the positive sentiment score and risk term density, the evaluation model can use the positive sentiment score, risk term density, and preset weights to calculate the brand friendliness score of the content to be evaluated. The specific calculation method is: Brand Friendliness = Positive Sentiment Score - Risk Term Density × Preset Weight. It can be seen that brand friendliness is positively correlated with the positive sentiment score and negatively correlated with the risk term density; that is, the higher the positive sentiment score, the higher the brand friendliness; the higher the risk term density, the lower the brand friendliness.

[0056] S230, Determine the content type and brand strategy of the content to be evaluated.

[0057] To ensure that the content output to be evaluated can play its greatest value in the most critical areas and thus serve the brand's overall strategy more efficiently, the evaluation system can first determine the content type (such as product reviews, news, tutorials) and brand strategy of the content to be evaluated.

[0058] S240 assigns corresponding weights to the scores of each quality indicator based on content type and brand strategy.

[0059] After determining the content type and brand strategy of the content to be evaluated, the evaluation system can flexibly adjust the evaluation criteria according to the content type and brand strategy to achieve the best results. The evaluation criteria can be represented by weights, so the evaluation system can assign corresponding weights to the scores of each quality indicator according to the content type and brand strategy, thereby achieving dynamic weight adjustment.

[0060] In one example, assuming the content to be evaluated is a product review, the content type is product review, and the brand strategy aims to persuade potential users through genuine experience and establish a professional and credible image, the evaluation system can dynamically adjust the weights as follows: High weighting: authenticity (evaluation data must be accurate), completeness (both advantages and disadvantages must be mentioned), and uniqueness of information (the evaluation perspective must be different from others). Medium weighting: readability (enabling readers to understand the testing process), brand friendliness (objectivity and neutrality are themselves a form of brand friendliness); Relatively low weight: Topic appeal (although it is important, sacrificing authenticity for the sake of attracting attention is not advisable).

[0061] In another example, assuming the content to be evaluated is a press release, the content type is news, and the brand strategy is to respond quickly to trending events, seize the initiative in public opinion, and increase brand exposure, the evaluation system can dynamically adjust the weights in the following way: High priority: authenticity (the lifeline of news), topicality (keeping up with hot topics), readability (quickly conveying core information); Medium weighting: unique information (can have a unique perspective) and brand friendliness (naturally integrate brand association in the report); Relatively low weight: completeness (news is fast-paced, but complete details can be added in subsequent in-depth reports).

[0062] In another example, suppose the content to be evaluated is a beauty tutorial video. The content type is tutorial, and the brand strategy is to help users build a professional and reliable expert image and cultivate user goodwill by providing valuable content. Then the evaluation system can dynamically adjust the weights in the following way: Extremely high weight: readability (clear steps, easy-to-understand language), completeness (from preparation to completion, every step is essential); High weight: Authenticity (the method must be effective and safe); Medium weighting: Brand friendliness (gently showcasing the use of the brand's products throughout the tutorial); Relatively low weight: uniqueness of information (standard tutorials do not have high requirements for uniqueness) and topicality (the core of the tutorial is practicality, not sparking a discussion).

[0063] S250 inputs the weights corresponding to the scores of each quality indicator into the evaluation model, and calculates the content quality score of the content to be evaluated by using the uniqueness score, topicality score, readability score, completeness score, authenticity score and brand friendliness score and their corresponding weights.

[0064] After allocating the weights w1 to w6 for uniqueness, w2 to w2 for topicality, w3 to w3 for readability, w4 to w4 for completeness, w5 to w5 for authenticity, and w6 to w6 for brand friendliness, the evaluation system can input the weights w1 to w6 for each quality indicator into the evaluation model. The evaluation model then uses the uniqueness, topicality, readability, completeness, authenticity, and brand friendliness scores, along with their corresponding weights w1 to w6, to calculate the content quality score of the content to be evaluated. The specific calculation formula is: Content Quality Score = Uniqueness Score × w1 + Topicality Score × w2 + Readability Score × w3 + Completeness Score × w4 + Authenticity Score × w5 + Brand Friendliness Score × w6.

[0065] S260 assesses the quality level of the content to be evaluated based on the content quality score.

[0066] This step can be found in the description in S130, and will not be repeated here.

[0067] In one implementation, assessing the quality level of the content to be evaluated based on a content quality score may include: If the content quality score is greater than or equal to the first threshold, and all quality indicator scores are greater than or equal to the second threshold, the quality level of the content to be evaluated is assessed as Level 1; or, if the content quality score is greater than or equal to the first threshold, and there are N quality indicator scores that are all less than the second threshold, the quality level of the content to be evaluated is assessed as Level 2, where N is a positive integer less than 6; or, if the content quality score is less than the first threshold, the quality level of the content to be evaluated is assessed as Level 3; where Level 1 > Level 2 > Level 3.

[0068] After calculating the content quality score and the score for each quality indicator, the evaluation system can compare the content quality score with a first threshold (e.g., 90 points) and compare the score for each quality indicator with a second threshold (e.g., 15 points) to assess the quality level of the content to be evaluated based on the two comparison results.

[0069] If the content quality score is greater than or equal to the first threshold (e.g., 90 points), and the score of each quality indicator is greater than or equal to the second threshold (e.g., 15 points), the evaluation system can assess the quality level of the content to be evaluated as Level 1. Level 1 is used to indicate that the content to be evaluated is of high quality, thus determining that the content to be evaluated is high-quality content. For high-quality content, the evaluation system can provide feedback on the approval result to users (e.g., content creators, operators). If the user clicks the "one-click publish" command, the evaluation system can transmit the high-quality content to the review workbench. After the review workbench verifies the compliance of the high-quality content, it is published to various social media platforms. The content recommendation systems of each social media platform recommend the high-quality content to relevant potential users.

[0070] If the content quality score is greater than or equal to the first threshold (e.g., 90 points), but there are N quality indicator scores (N=1~5) that are all less than the second threshold (e.g., 15 points), the evaluation system can assess the quality level of the content to be evaluated as Level 2. Level 2 is used to represent that the content quality of the content to be evaluated is medium, so the content to be evaluated can be determined as qualified content. For qualified content, the evaluation system can provide users (e.g., content creators, operators) with optimization suggestions for the corresponding content items. Optimization suggestions include suggestions for the dimensions where the quality indicator scores are less than the second threshold.

[0071] If the content quality score is less than the first threshold (e.g., 90 points), the evaluation system can classify the content as level three. Level three indicates low content quality, thus identifying the content as poor quality. For poor quality content, the evaluation system can provide optimization suggestions for various content items to users (e.g., content creators, operators). These suggestions include optimization recommendations for dimensions where the quality indicator score is less than the second threshold.

[0072] It should be noted that the evaluation system can output a diagnostic report, including the content quality score, the score for each quality indicator, and the results or optimization suggestions, and then present the diagnostic report to users (such as content creators and operators) in a visual format.

[0073] It is evident that the embodiments of this application differ significantly from related technologies in terms of processing efficiency. Related technologies primarily rely on manual review for content quality assessment; however, manual review is limited to 10-20 items per person per hour, resulting in low efficiency, high cost, and difficulty in scaling. In contrast, the embodiments of this application are fully automated with millisecond-level response times. For example, the assessment system can process thousands of items per second, thereby greatly improving content quality assessment efficiency—for instance, by more than 300 times—and eliminating the need for a large number of reviewers, thus reducing labor costs by up to 90%.

[0074] Furthermore, the embodiments of this application differ significantly from related technologies in terms of evaluation consistency: related technologies rely heavily on individual judgment (such as the experience of auditors), which can easily lead to inconsistent evaluation standards due to human subjectivity, resulting in uneven content quality. In contrast, the embodiments of this application are entirely data- and algorithm-driven, which can unify evaluation standards and achieve 99.9% consistency in evaluation results, completely eliminating human subjective bias and fatigue errors.

[0075] Furthermore, the embodiments of this application differ significantly from related technologies in terms of coverage dimensions: related technologies partially rely on automated review methods for content quality assessment; however, this method only covers a single dimension (such as keyword filtering) and cannot comprehensively measure content value. In contrast, the embodiments of this application cover a multi-dimensional indicator encompassing six core dimensions: uniqueness, topicality, readability, completeness, authenticity, and brand friendliness. Therefore, this multi-dimensional indicator can be used to comprehensively measure content value, greatly improving the accuracy of content quality assessment. For example, it reduces the misjudgment rate of content value (high-quality content mistakenly deleted / low-quality content overlooked) by 70%.

[0076] Furthermore, the embodiments of this application differ significantly from related technologies in terms of business value: related technologies rely on manual or automated review methods for content quality assessment; however, both methods are characterized by lag and passive defense. In contrast, the embodiments of this application are forward-looking and proactively empowering. For example, after a content recommendation system adopts the content quality score provided by the embodiments of this application, the click-through rate of high-quality content (content quality score ≥ 90 points) can increase by 35%, and the average dwell time of potential users can increase by 20%.

[0077] Corresponding to the aforementioned application function implementation method embodiments, this application also provides a content quality assessment system. Unification Electronic devices and corresponding embodiments.

[0078] Figure 3 This is a schematic diagram of the content quality assessment system shown in the embodiments of this application.

[0079] See Figure 3 The content quality assessment system of this application may include: The acquisition module 310 is used to acquire the content to be evaluated and call the pre-trained evaluation model; the evaluation model is trained using multiple historical published contents. The calculation module 320 is used to input the content to be evaluated into the evaluation model so as to calculate the content quality score of the content to be evaluated through the evaluation model; Evaluation module 330 is used to evaluate the quality level of the content to be evaluated based on the content quality score.

[0080] In one embodiment, the computing module 320 may include: The Quality Indicator Score Calculation Submodule is used to calculate the quality indicator score of the content to be evaluated through the evaluation model; the quality indicator score includes at least one of the following: uniqueness score, topicality score, readability score, completeness score, authenticity score, and brand friendliness score; The attribute determination submodule is used to determine the content type and brand strategy of the content to be evaluated; The weight allocation submodule is used to assign corresponding weights to the scores of various quality indicators based on content type and brand strategy; The content quality score calculation submodule is used to input the weights corresponding to the scores of various quality indicators into the evaluation model. The evaluation model uses uniqueness score, topicality score, readability score, completeness score, authenticity score, brand friendliness score and corresponding weights to calculate the content quality score of the content to be evaluated.

[0081] In one embodiment, the evaluation module 330 may include: The first-level assessment submodule is used to assess the quality level of the content to be evaluated as first-level if the content quality score is greater than or equal to the first threshold and all quality indicator scores are greater than or equal to the second threshold; or, The second-level assessment submodule is used to assess the quality level of the content to be evaluated as the second level if the content quality score is greater than or equal to the first threshold, and there are N quality indicator scores that are all less than the second threshold, where N is a positive integer less than 6; or, The third-level assessment submodule is used to assess the quality level of the content to be evaluated as the third level if the content quality score is less than the first threshold. The order is: First level > Second level > Third level.

[0082] In one embodiment, the quality index score calculation submodule may include: The similarity calculation unit is used to calculate the similarity between the content to be evaluated and each historically published content through the evaluation model. The descending sorting unit is used to sort multiple similarities in descending order by the evaluation model in order to select the target similarity that is first; The uniqueness score calculation unit is used to calculate the uniqueness score of the content to be evaluated based on the target similarity through the evaluation model; the uniqueness score is negatively correlated with the target similarity.

[0083] In one embodiment, the quality index score calculation submodule may include: The feature extraction unit is used to extract the popularity features, timeliness features, and interaction potential features of the content to be evaluated through the evaluation model; The topicality score calculation unit is used to calculate the topicality score of the content to be evaluated based on the popularity feature, timeliness feature, and interaction potential feature through the evaluation model; wherein, the topicality score is positively correlated with the popularity feature, timeliness feature, and interaction potential feature.

[0084] In one embodiment, the quality index score calculation submodule may include: A reading friendliness determination unit is used to determine the reading friendliness of the content to be evaluated based on the average sentence length and average word length of the content to be evaluated using an evaluation model; and, The narrative structure completeness determination unit is used to determine the narrative structure completeness of the content to be evaluated based on the chapter markers of the content to be evaluated through the evaluation model. The readability score calculation unit is used to calculate the readability score of the content to be evaluated based on reading friendliness and narrative structure completeness through an evaluation model; the readability score is positively correlated with reading friendliness and narrative structure completeness.

[0085] In one embodiment, the quality index score calculation submodule may include: The element completeness calculation unit is used to calculate the element completeness of the content to be evaluated by using the evaluation model; different content types correspond to different preset key elements. The integrity score calculation unit is used to calculate the integrity score of the content to be evaluated based on the element completeness through the evaluation model; the integrity score is positively correlated with the element completeness.

[0086] In one embodiment, the quality index score calculation submodule may include: The fact statement extraction unit is used to extract fact statements from the content to be evaluated through the evaluation model. The verification unit is used to verify factual statements using an official database through an evaluation model and obtain verification results. The authenticity score calculation unit is used to calculate the authenticity score of the content to be evaluated based on the verification results using the evaluation model.

[0087] In one embodiment, the quality index score calculation submodule may include: The sentiment analysis unit is used to perform sentiment analysis on the content to be evaluated using an evaluation model, thereby obtaining a positive sentiment score for the content to be evaluated; and, The risk word density calculation unit is used to calculate the risk word density of the content to be evaluated containing preset risk words through the evaluation model. The brand friendliness score calculation unit is used to calculate the brand friendliness score of the content to be evaluated based on the positive sentiment score and risk word density through the evaluation model. Among them, brand friendliness is positively correlated with the positive sentiment score, and brand friendliness is negatively correlated with risk word density.

[0088] As this example illustrates, the solution provided in this application involves acquiring the content to be evaluated and calling a pre-trained evaluation model. The evaluation model is trained using multiple historically published content samples. The content to be evaluated is input into the evaluation model to calculate its content quality score. Based on this score, the quality level of the content is assessed. This application calculates the content quality score using an evaluation model. The content quality score effectively measures the value of the content itself, thus accurately assessing its quality level. The entire process is automated, with millisecond-level response, and is entirely data- and algorithm-driven, thereby improving the efficiency and accuracy of content quality assessment, reducing labor costs, and scalability.

[0089] Regarding the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated further here.

[0090] Figure 4 This is a schematic diagram of the structure of an electronic device shown in an embodiment of this application.

[0091] See Figure 4 The electronic device 400 includes a memory 410 and a processor 420.

[0092] The processor 420 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. Memory 410 may include various types of storage units, such as system memory, read-only memory (ROM), and permanent storage devices. ROM may store static data or instructions required by processor 420 or other modules of the computer. Permanent storage devices may be read-write storage devices. Permanent storage devices may be non-volatile storage devices that retain stored instructions and data even when the computer is powered off. In some embodiments, permanent storage devices use mass storage devices (e.g., magnetic or optical disks, flash memory) as permanent storage devices. In other embodiments, permanent storage devices may be removable storage devices (e.g., floppy disks, optical drives). System memory may be a read-write storage device or a volatile read-write storage device, such as dynamic random access memory. System memory may store some or all of the instructions and data required by the processor during operation. Furthermore, memory 410 may include any combination of computer-readable storage media, including various types of semiconductor memory chips (e.g., DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), and disks and / or optical disks may also be used. In some embodiments, memory 410 may include a removable storage device that is readable and / or writable, such as a laser disc (CD), a read-only digital multifunction optical disc (e.g., DVD-ROM, dual-layer DVD-ROM), a read-only Blu-ray disc, an ultra-high density optical disc, a flash memory card (e.g., SD card, mini SD card, Micro-SD card, etc.), a magnetic floppy disk, etc. Computer-readable storage media do not contain carrier waves or transient electronic signals transmitted wirelessly or via wired connections.

[0093] The memory 410 stores executable code, which, when processed by the processor 420, can cause the processor 420 to execute part or all of the methods described above.

[0094] Furthermore, the method according to this application can also be implemented as a computer program or computer program product, which includes computer program code instructions for performing some or all of the steps in the method described above.

[0095] Alternatively, this application may be implemented as a computer-readable storage medium (or a non-transitory machine-readable storage medium or a machine-readable storage medium) storing executable code (or computer program or computer instruction code) that, when executed by a processor of an electronic device (or server, etc.), causes the processor to perform part or all of the steps of the methods described above according to this application.

[0096] This application also provides a computer program product, which includes computer instructions that, when executed by a processor, implement the method described above.

[0097] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A content quality assessment method, characterized in that, include: The content to be evaluated is obtained, and a pre-trained evaluation model is invoked; the evaluation model is trained using multiple historical published contents. The content to be evaluated is input into the evaluation model to calculate the content quality score of the content to be evaluated. The quality level of the content to be evaluated is determined based on the content quality score.

2. The method according to claim 1, characterized in that, The calculation of the content quality score of the content to be evaluated using the evaluation model includes: The quality index score of the content to be evaluated is calculated using the evaluation model; the quality index score includes at least one of the following: uniqueness score, topicality score, readability score, completeness score, authenticity score, and brand friendliness score. Determine the content type and brand strategy of the content to be evaluated; Based on the content type and the brand strategy, assign corresponding weights to the scores of each of the quality indicators; The weights corresponding to the scores of each of the aforementioned quality indicators are input into the evaluation model, so that the evaluation model can calculate the content quality score of the content to be evaluated using the uniqueness score, the topicality score, the readability score, the completeness score, the authenticity score, the brand friendliness score, and the corresponding weights.

3. The method according to claim 2, characterized in that, The process of evaluating the quality level of the content to be evaluated based on the content quality score includes: If the content quality score is greater than or equal to the first threshold, and all the scores of the quality indicators are greater than or equal to the second threshold, then the quality level of the content to be evaluated is assessed as the first level; or, If the content quality score is greater than or equal to the first threshold, and there are N quality indicator scores that are all less than the second threshold, then the quality level of the content to be evaluated is assessed as the second level, where N is a positive integer less than 6; or, If the content quality score is less than the first threshold, the quality level of the content to be evaluated will be assessed as the third level. Among them, the first level > the second level > the third level.

4. The method according to claim 2, characterized in that, The calculation of the quality index score of the content to be evaluated using the evaluation model includes: The similarity between the content to be evaluated and each of the historical published contents is calculated using the evaluation model. The evaluation model is used to sort the multiple similarities in descending order to select the target similarity that is first. The evaluation model calculates the uniqueness score of the content to be evaluated based on the target similarity; wherein the uniqueness score is negatively correlated with the target similarity.

5. The method according to claim 2, characterized in that, The calculation of the quality index score of the content to be evaluated using the evaluation model includes: The evaluation model is used to extract the popularity characteristics, timeliness characteristics, and interaction potential characteristics of the content to be evaluated. The evaluation model calculates the topicality score of the content to be evaluated based on the popularity feature, the timeliness feature, and the interaction potential feature; wherein the topicality score is positively correlated with the popularity feature, the timeliness feature, and the interaction potential feature.

6. The method according to claim 2, characterized in that, The calculation of the quality index score of the content to be evaluated using the evaluation model includes: The evaluation model determines the readability of the content to be evaluated based on the average sentence length and average word length; and, The narrative structure completeness of the content to be evaluated is determined by the evaluation model based on the chapter markers of the content to be evaluated. The evaluation model calculates the readability score of the content to be evaluated based on the reading friendliness and the narrative structure completeness; wherein the readability score is positively correlated with the reading friendliness and the narrative structure completeness.

7. The method according to claim 2, characterized in that, The calculation of the quality index score of the content to be evaluated using the evaluation model includes: The evaluation model is used to calculate the element completeness of the content to be evaluated, which contains preset key elements; wherein, different content types correspond to different preset key elements. The evaluation model calculates the completeness score of the content to be evaluated based on the element completeness; the completeness score is positively correlated with the element completeness.

8. The method according to claim 2, characterized in that, The calculation of the quality index score of the content to be evaluated using the evaluation model includes: The factual statements in the content to be evaluated are extracted using the evaluation model. The factual statements are verified using the official database through the evaluation model, and the verification results are obtained. The evaluation model calculates the authenticity score of the content to be evaluated based on the verification results.

9. The method according to claim 2, characterized in that, The calculation of the quality index score of the content to be evaluated using the evaluation model includes: Sentiment analysis is performed on the content to be evaluated using the evaluation model to obtain a positive sentiment score for the content to be evaluated; and, The risk word density of the content to be evaluated containing preset risk words is calculated using the evaluation model. The evaluation model calculates the brand friendliness score of the content to be evaluated based on the positive sentiment score and the risk word density; wherein the brand friendliness is positively correlated with the positive sentiment score and negatively correlated with the risk word density.

10. A content quality assessment system, characterized in that, include: The acquisition module is used to acquire the content to be evaluated and call a pre-trained evaluation model; the evaluation model is trained using multiple historical published contents. The calculation module is used to input the content to be evaluated into the evaluation model, so as to calculate the content quality score of the content to be evaluated through the evaluation model; The evaluation module is used to evaluate the quality level of the content to be evaluated based on the content quality score.

Images