CN114428837A - Content quality evaluation method, device, medium and electronic equipment - Google Patents

Abstract

The present disclosure relates to a content quality evaluation method, device, medium and electronic device. The method includes: obtaining a first evaluation result and evaluation text of a target user for target content; and determining a first evaluation result of the target content according to the evaluation text. Second evaluation result; determine the matching degree corresponding to the target user and the target content; according to the first evaluation result, the second evaluation result and the matching degree, determine the target user's degree of matching with respect to the target content Quality evaluation results. Therefore, the determination of the quality evaluation result in combination with the above three can fully consider the evaluation score and the comment text input by the user, so as to maintain consistency between the determined quality evaluation result and the evaluation text input by the user. At the same time, combining the matching degree between the user and the content can reduce the subjective influence of the user to a certain extent, ensure the accuracy and objectivity of the quality evaluation result, and improve the matching degree between the quality evaluation result and the target content.

Description

Content quality evaluation method, device, medium and electronic equipment

technical field

The present disclosure relates to the field of data processing, and in particular, to a content quality evaluation method, apparatus, medium and electronic device.

Background technique

Text sentiment analysis refers to the process of analyzing, processing and extracting subjective texts with emotional color by using natural language processing technology, text mining and computer linguistics.

For content quality evaluation systems such as movies and TV series, audience users will give their own evaluation texts and evaluation scores for the content, so as to realize the quality evaluation of the content and recommend corresponding content to other users. However, in the above evaluation methods, the rating is usually based on a five-point system. When the user directly enters the rating, it is easy to directly give a full-point evaluation. However, the evaluation text corresponding to the user may not be completely positive, that is, to a certain extent, the evaluation input by the user Ratings and evaluation texts often do not match, so that the evaluation scores of many contents are often falsely high, making it difficult to provide users with accurate content viewing references.

SUMMARY OF THE INVENTION

The purpose of the present disclosure is to provide an accurate and objective content quality evaluation method, device, medium and electronic device.

In order to achieve the above object, according to a first aspect of the present disclosure, a content quality evaluation method is provided, the method comprising:

Obtain the first evaluation result and evaluation text of the target user for the target content;

determining a second evaluation result of the target content according to the evaluation text;

determining the matching degree corresponding to the target user and the target content;

According to the first evaluation result, the second evaluation result and the matching degree, the quality evaluation result of the target user for the target content is determined.

Optionally, the determining the matching degree corresponding to the target user and the target content includes:

Determine the user vector corresponding to the target user based on the historical evaluation text of the target user;

determining a content vector corresponding to the target content based on a plurality of evaluation texts corresponding to the target content;

The matching degree is determined according to the user vector and the content vector.

Optionally, determining the user vector corresponding to the target user based on the historical evaluation text of the target user, including:

Clustering the historical evaluation text to obtain a plurality of clusters corresponding to the historical evaluation text;

For each of the clusters, splicing historical evaluation texts whose text length is less than a preset length threshold in the cluster to obtain at least one spliced text;

Based on the spliced text in each cluster, the historical evaluation text whose text length is not less than the length threshold, and the topic generation model, determine the subject word corresponding to the target user, and based on the vector corresponding to the subject word, determine the user vector.

Optionally, determining the content vector corresponding to the target content based on multiple evaluation texts corresponding to the target content, including:

Determine the word frequency and reverse document frequency corresponding to each word segment in the multiple evaluation texts corresponding to the target content, and the text length ratio corresponding to the word segment, wherein the text length ratio corresponding to the word segment is the segment to which the word segment belongs. the ratio of the length of the evaluation text to the average text length of the plurality of evaluation texts;

For each of the word segmentations, the product of the word frequency corresponding to the word segmentation, the reverse document frequency and the text length ratio corresponding to the word segmentation is determined as the target parameter corresponding to the word segmentation;

According to the target parameter corresponding to each of the word segments, the target word segment corresponding to the target content is determined, and the content vector is determined based on the vector corresponding to the target word segment.

Optionally, the determining the second evaluation result of the target content according to the evaluation text includes:

According to the evaluation text and the text classification model, the classification corresponding to the evaluation text is determined, and the score indicated by the classification is determined as the second evaluation result;

Wherein, in the training process of the text classification model, feature extraction is performed based on the feature extraction sub-model and target features are obtained based on the fully connected layer, and the prediction result of the text classification model is obtained by predicting some features in the target features .

Optionally, determining the quality evaluation result of the target user for the target content according to the first evaluation result, the second evaluation result and the matching degree includes:

determining the weighted sum of the first evaluation result and the second evaluation result as the initial evaluation result;

The initial evaluation result is adjusted according to the matching degree to obtain the quality evaluation result.

Optionally, the method further includes:

Determine the recommended content corresponding to the target user according to the quality evaluation result corresponding to each content in the content library;

The recommended content is output.

According to a second aspect of the present disclosure, there is provided an apparatus for evaluating content quality, the apparatus comprising:

an acquisition module, used to acquire the first evaluation result and evaluation text of the target user for the target content;

a first determining module, configured to determine a second evaluation result of the target content according to the evaluation text;

a second determining module, configured to determine the matching degree corresponding to the target user and the target content;

The third determination module is configured to determine the quality evaluation result of the target user for the target content according to the first evaluation result, the second evaluation result and the matching degree.

According to a third aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium having a computer program stored thereon, the program implementing the steps of any one of the methods in the first aspect when the program is executed by a processor.

According to a fourth aspect of the present disclosure, there is provided an electronic device, comprising:

a memory on which a computer program is stored;

A processor, configured to execute the computer program in the memory, to implement the steps of any one of the methods of the first aspect.

Therefore, through the above technical solution, the first evaluation result is the score input by the target user, the second evaluation result is the score determined according to the evaluation text input by the target user, and the matching degree corresponding to the target user and the target content can be to a certain extent Indicates the possibility of subjective evaluation by the user, so that the quality evaluation result determined by combining the above three can fully consider the evaluation score and comment text input by the user, so as to maintain consistency between the determined quality evaluation result and the evaluation text input by the user . At the same time, combining the matching degree between the user and the content can reduce the subjective influence of the user to a certain extent, so as to ensure the accuracy and objectivity of the quality evaluation result, and improve the matching degree between the quality evaluation result and the target content. Users can provide accurate content viewing reference to improve user experience.

Other features and advantages of the present disclosure will be described in detail in the detailed description that follows.

Description of drawings

The accompanying drawings are used to provide a further understanding of the present disclosure, and constitute a part of the specification, and together with the following detailed description, are used to explain the present disclosure, but not to limit the present disclosure. In the attached image:

1 is a flowchart of a content quality evaluation method provided according to an embodiment of the present disclosure;

2 is a schematic structural diagram of a text classification model provided according to an embodiment of the present disclosure;

3 is a flowchart of an exemplary implementation of determining a corresponding degree of match between a target user and target content;

FIG. 4 is a block diagram of a content quality evaluation apparatus provided according to an embodiment of the present disclosure;

5 is a block diagram of an electronic device according to an exemplary embodiment;

Fig. 6 is a block diagram of an electronic device according to an exemplary embodiment.

Detailed ways

The specific embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present disclosure, but not to limit the present disclosure.

FIG. 1 is a flowchart of a content quality evaluation method provided according to an embodiment of the present disclosure. As shown in Figure 1, the method may include:

In step 11, the first evaluation result and evaluation text of the target user for the target content are obtained. Wherein, the target content may be any content in multimedia content such as movies, TV series, and animations. For example, the first evaluation result may be a rating value input by the target user for the target content, and the evaluation text may be comment characters input by the target user for the target content.

In step 12, a second evaluation result of the target content is determined according to the evaluation text.

The evaluation text is the target user's comment on the target content, so the evaluation text can also reflect the evaluation bias of the target user on the target content to a certain extent. Therefore, in this embodiment, it can be further determined according to the evaluation text. Another evaluation result of the target user on the target content. For example, the second evaluation result can be represented by a score value, and the value range of the score value is the same as the range of the score value corresponding to the first evaluation result, then the second evaluation result can be used to indicate that according to the target user input The rating determined by the review text.

In step 13, the matching degree corresponding to the target user and the target content is determined.

Among them, each target user may have its corresponding field of interest, which belongs to the target content in the field of interest, and the user may give a better evaluation, that is, the user may give a biased evaluation. Therefore, in this embodiment, the matching degree between the target user and the target content can be determined to represent the possibility of subjective evaluation of the user to a certain extent, that is, the higher the matching degree, the more subjective the evaluation given by the user. The higher the probability of sexual evaluation.

In step 14, according to the first evaluation result, the second evaluation result and the matching degree, the quality evaluation result of the target user for the target content is determined.

Therefore, through the above technical solution, the first evaluation result is the score input by the target user, the second evaluation result is the score determined according to the evaluation text input by the target user, and the matching degree corresponding to the target user and the target content can be to a certain extent Indicates the possibility of subjective evaluation by the user, so that the quality evaluation result determined by combining the above three can fully consider the evaluation score and comment text input by the user, so as to maintain consistency between the determined quality evaluation result and the evaluation text input by the user . At the same time, combining the matching degree between the user and the content can reduce the subjective influence of the user to a certain extent, so as to ensure the accuracy and objectivity of the quality evaluation result, and improve the matching degree between the quality evaluation result and the target content. Users can provide accurate content viewing reference to improve user experience.

As an example, the rating text may be predicted based on the rating model to obtain the second rating result. For example, the scoring model can be implemented based on CNN (Convolutional Neural Networks, Convolutional Neural Networks), RNN (Recurrent Neural Networks, Recurrent Neural Networks), etc., and the evaluation texts based on annotation scores are used as training samples for training, and the training method can use this method. The training is carried out in a common way in the field, and will not be repeated here.

As another example, an exemplary implementation manner of determining the second evaluation result of the target content according to the evaluation text in step 12 is as follows, including:

According to the evaluation text and the text classification model, the classification corresponding to the evaluation text is determined, and the score indicated by the classification is determined as the second evaluation result.

Wherein, in the training process of the text classification model, feature extraction is performed based on the feature extraction sub-model and target features are obtained based on the fully connected layer, and the prediction result of the text classification model is obtained by predicting some features in the target features .

Exemplarily, the feature extraction sub-model can be implemented based on the Transformer model. Correspondingly, a schematic diagram of the structure of the text classification model is shown in FIG. 2 , wherein the number of layers of each feature layer in the text classification model is only an exemplary illustration, not the Public plans are limited.

As shown in Figure 2, the text classification model includes a feature extraction sub-model, which can be composed of a Transformer model, and then includes a fully-connected layer, a fully-connected layer, a random deactivation layer, DropoutLayer, and can be combined with excitation functions and normalization. layer to get the prediction result. Illustratively, the excitation function may be GELU. For example, if the set score is 1-5 points, 5 categories can be set in the normalization layer, corresponding to 1, 2, 3, 4, and 5 points respectively.

Therefore, in this embodiment, the evaluation text can be segmented to obtain a segmented sequence corresponding to the evaluation text, and then each segmented word can be vectorized to represent the input text classification model. In the text classification model, the feature extraction can be based on attention The mechanism pays more attention to the features related to scoring. After extracting the features, the target features are obtained based on the fully connected layer to ensure the accuracy of feature extraction and provide reliable data support for subsequent predictions. Moreover, the prediction result of the text classification model is obtained by predicting some features in the target features. For example, the probability of dropout of each neural network layer can be set through the Dropout Layer. For the neural network training unit, according to this probability, It is removed from the network for prediction classification to obtain predictions and train them. Therefore, during the training process of the text classification model, prediction can be made randomly based on some features in the target features, and for stochastic gradient descent, part of the features are selected randomly, so that every feature of the text classification model is selected randomly. A mini-batch is training different networks, which can effectively prevent the training of the text classification model from entering into overfitting, thereby improving the accuracy of the text classification model obtained by training, and ensuring the accuracy of the determined second evaluation result. The content is objectively and accurately evaluated to provide support.

In a possible embodiment, an exemplary implementation manner of determining the corresponding degree of matching between the target user and the target content in step 13 is as follows. As shown in FIG. 3 , this step may include:

In step 31, a user vector corresponding to the target user is determined based on the historical evaluation text of the target user.

Wherein, in this step, each evaluation text of the target user may be obtained under the condition of obtaining the user's authorization, so that the characteristics of the user may be obtained from the historical evaluation of the user, that is, the user vector may be obtained.

As an example, keyword extraction may be performed based on multiple historical evaluation texts of the target user, so that the extracted keyword may be used as the characteristic word of the target user, and then the user vector may be obtained by vectorizing the characteristic word.

As another example, the determining the user vector corresponding to the target user based on the historical evaluation text of the target user may include:

The historical evaluation text is clustered to obtain a plurality of clusters corresponding to the historical evaluation text.

Among them, the historical evaluation text can be clustered based on a common clustering algorithm such as K-means or KNN to obtain multiple clusters.

For each of the clusters, the historical evaluation texts whose text lengths are less than a preset length threshold in the cluster are spliced to obtain at least one spliced text.

Among them, when the user evaluates the content, the evaluation may be performed frequently through short texts, and if the length of the evaluation text is too short, it is inconvenient to generate the subject words. Therefore, in this embodiment, the historical evaluation texts may be clustered first, so that historical evaluation texts with similar characteristics are clustered together.

For example, the preset length threshold may be set based on an actual application scenario, which is not limited in the present disclosure. For each cluster, if the text length of the historical evaluation text is less than the length threshold, it means that the historical evaluation text is short text. The texts with similar features extend the length of the historical evaluation texts, and at the same time to realize the text expansion, the own characteristics of each historical evaluation text are guaranteed.

As an example, all historical evaluation texts whose text lengths are less than a preset length threshold in a cluster may be spliced to obtain a spliced text. As another example, in order to avoid excessively long spliced text, the text length limit of the spliced text can be set, that is, when splicing based on the short text in the current cluster, the text length of the current spliced text S is less than the text length Limit, if the text length of the spliced text S' obtained after adding a new historical evaluation text A is greater than the text length limit, the splicing will not be performed, that is, the current spliced text S is regarded as the spliced text that has been spliced, and the The newly added historical evaluation text A is used as a new spliced text to be further spliced with other historical evaluation texts to obtain multiple spliced texts.

After that, based on the spliced text in each cluster, the historical evaluation text whose text length is not less than the length threshold, and the topic generation model, determine the subject word corresponding to the target user, and based on the vector corresponding to the subject word , determine the user vector.

Wherein, the text length of the historical evaluation text is not less than the length threshold, indicating that the historical evaluation text can be directly applied to the subject word generation. As described above, the spliced text is obtained by splicing the historical evaluation text whose text length is less than the length threshold in each cluster, so that the spliced text is a text with a longer text length. Therefore, in this embodiment, the subject words can be obtained based on the subject generation model based on the spliced text and the historical evaluation texts that have not been spliced in the cluster. Set the number of words as subject words, and then vectorize the subject words to obtain the user vector used to characterize the user's characteristics.

For example, the topic generation model may be implemented based on LDA (Latent Dirichlet Allocation, latent Dirichlet distribution), and the training method thereof is the prior art, which will not be repeated here. The LDA model is an unsupervised model, and in this embodiment, the LDA model can be pre-trained based on existing prior texts. Content evaluation text, etc., so that to a certain extent, the topic generation model can improve the topic word determination in customized scenarios that users pay attention to.

In this way, texts with similar characteristics and shorter text lengths can be combined and spliced, so that the subject words can be determined based on the long texts, and the accuracy of the subject words determination can be improved to a certain extent, that is, the accuracy of the determined user characteristics can be improved. Accuracy, providing accurate data support for determining user portraits.

In step 32, a content vector corresponding to the target content is determined based on a plurality of evaluation texts corresponding to the target content.

Among them, multiple users can evaluate the same content. Therefore, based on the evaluation texts of multiple users on the same target content, the real comprehensive feature of the target content can be obtained, that is, the content vector can be obtained.

As an example, keyword extraction may be performed based on multiple evaluation texts, so that the extracted keywords may be used as feature words of the target content, and then the content vector may be obtained by vectorizing the feature words.

As another example, the determining a content vector corresponding to the target content based on multiple evaluation texts corresponding to the target content may include:

Determine the word frequency and reverse document frequency corresponding to each word segment in the multiple evaluation texts corresponding to the target content, and the text length ratio corresponding to the word segment, wherein the text length ratio corresponding to the word segment is the segment to which the word segment belongs. The ratio of the length of the evaluation text to the average text length of the plurality of evaluation texts.

Wherein, each evaluation text can be regarded as a document, and the evaluation text can be segmented to obtain each segment corresponding to the multiple evaluation texts. Thus, the word frequency TF, inverse document frequency IDF and text length ratio DOC_LEN corresponding to each word segment can be further determined. The formula is as follows:

For each of the word segmentations, the product of the word frequency corresponding to the word segmentation, the reverse document frequency and the text length ratio corresponding to the word segmentation is determined as the target parameter corresponding to the word segmentation; according to the target parameter corresponding to each of the word segmentation, A target word segment corresponding to the target content is determined, and the content vector is determined based on a vector corresponding to the target word segment.

Among them, the top N word segmentations ranked according to the target parameters corresponding to the word segmentation in descending order can be determined as the target word segmentation corresponding to the target content, and then the target word segmentation can be vectorized to obtain a content vector. Among them, the vectorization of the subject words of the target user is the same as the vectorization of the target word segmentation of the target content, and the vectorization method commonly used in the field can be selected, such as the word2vec method, which is not limited in this disclosure. .

Therefore, when segmenting the target word from the evaluation text, the word frequency of the segmented word and the reverse document frequency are considered, and the text length of the evaluation text to which the segmented word belongs is also combined. As shown above, it is difficult to accurately extract keywords from evaluation texts with short text lengths. Therefore, in the present disclosure, the importance of word segmentation in evaluation texts with longer text lengths is increased by considering the text length ratio, so as to ensure the accuracy of the determined target word segmentations. accuracy, thereby improving the accuracy and comprehensiveness of the features in the content vector.

In step 33, the matching degree is determined according to the user vector and the content vector.

Exemplarily, the cosine similarity between the user vector and the content vector can be calculated based on the user vector and the content vector. As the matching degree, for example, the following formula can be used to calculate:

Among them, β ₁ is used to represent the user vector, β ₂ is used to represent the content vector, Ai is used to represent the ith feature in the user vector, Bi is used to represent the ith feature in the content vector, and n is used to represent the user vector and the dimension of the feature of the content vector, the number of dimensions corresponding to the two is the same.

Therefore, through the above technical solution, the interest characteristics of the target user can be obtained based on the user's historical evaluation texts, and the characteristics of the target content itself can be obtained based on multiple evaluation texts of the target content, so that the user can be characterized based on the matching degree between the two. Whether an objective evaluation of the target content can be given to provide data parameters for the subsequent determination of the final quality evaluation result based on the matching degree, so as to ensure the objectivity of the determined quality evaluation result.

In a possible embodiment, in step 14, according to the first evaluation result, the second evaluation result and the matching degree, an exemplary implementation manner of determining the quality evaluation result of the target user for the target content is as follows, and this step may include:

A weighted sum of the first evaluation result and the second evaluation result is determined as an initial evaluation result.

For example, the respective corresponding weights of the first evaluation result and the second evaluation result may be preset, wherein the sum of the respective weights corresponding to the first evaluation result and the second evaluation result is 1, and the respective corresponding weights may be determined according to the actual situation. The application scenario is set, which is not limited in this disclosure. As an example, the weights corresponding to the two are 0.5, and the average value of the first evaluation result and the second evaluation result can be determined as the initial evaluation result, so as to comprehensively consider the consistency between the score input by the user and the text sex.

The initial evaluation result is adjusted according to the matching degree to obtain the quality evaluation result.

Among them, as mentioned above, the matching degree is used to indicate the subjectivity degree of the target user's evaluation of the target content. Therefore, when it is determined that the quality evaluation result needs to offset the influence of subjective preference factors, the initial evaluation result is adjusted according to the matching degree, and the result obtained by subtracting the matching degree from the initial evaluation result can be used as The quality evaluation results. Alternatively, according to the preset adjustment weight of the matching degree, the product of the matching degree and the adjustment weight of the matching degree may be used as the adjustment value, and the result obtained by subtracting the adjustment value from the initial evaluation result may be used as the quality evaluation result.

Therefore, through the above technical solution, the determined quality evaluation result is consistent with the evaluation score and evaluation text input by the user, and at the same time, the influence of the user's subjective preference can be reduced to a certain extent, so as to ensure the accuracy of the determined quality evaluation result. sex and objectivity. In addition, to a certain extent, the quality evaluation results are more dispersed compared with the first evaluation results input by the user, so as to avoid the situation where the evaluation results of the content overlap greatly and it is difficult to distinguish the quality of the content, and clarify each content for the user. The actual quality of the data provides a more accurate data reference.

In a possible embodiment, the method may further include:

According to the quality evaluation result corresponding to each content in the content library, the recommended content corresponding to the target user is determined; the recommended content is output. Wherein, the quality evaluation result corresponding to all or part of the content in the content library is determined according to the content quality evaluation method described above.

As an example, the top P may be selected as the recommended content for the target user according to the quality evaluation result corresponding to each content in the content library in descending order, where P may be set according to actual user needs. As a result, higher-quality content can be recommended for target users, and the user's content viewing experience can be guaranteed.

As another example, the top Q may be selected as the candidate content of the target user according to the quality evaluation result corresponding to each content in the content library in descending order, where Q may be set according to actual user needs. After that, based on the selection matching degree between the candidate content and the target user, the top P is selected as the recommended content in the order of the selection matching degree from high to low and displayed in the order of the selection matching degree from high to low, where P is less than or equal to Q. The selection matching degree may be determined by similarity calculation based on the tag vector of the candidate content and the interest vector of the target user, so as to indicate whether the candidate content conforms to the user's interest preference. Illustratively, the label vector may be obtained based on the type label vectorization of the candidate content, and the type label may be a genre (such as a feature film, a documentary), a genre (such as a comedy, tragedy, urban, rural), etc., and the interest vector may be It is obtained from the vectorization of the user's interest tag obtained under the authorization of the user, and the interest tag can be idol drama, youth, city, etc. In this way, while recommending high-quality content for the target user, the recommendation display can be performed based on the target user's interests and preferences, so as to ensure the matching degree between the recommended content and the target user, and at the same time improve the diversity and individuality of the recommended content to further enhance the user experience.

Based on the same inventive concept, the present disclosure also provides an apparatus for evaluating content quality. As shown in FIG. 4 , the apparatus 10 includes:

The obtaining module 100 is used for obtaining the first evaluation result and evaluation text of the target user for the target content;

a first determination module 200, configured to determine a second evaluation result of the target content according to the evaluation text;

a second determining module 300, configured to determine the matching degree corresponding to the target user and the target content;

The third determination module 400 is configured to determine, according to the first evaluation result, the second evaluation result and the matching degree, the quality evaluation result of the target user for the target content.

Optionally, the second determining module includes:

a first determination submodule, configured to determine a user vector corresponding to the target user based on the historical evaluation text of the target user;

a second determination submodule, configured to determine a content vector corresponding to the target content based on a plurality of evaluation texts corresponding to the target content;

The third determining submodule is configured to determine the matching degree according to the user vector and the content vector.

Optionally, the first determination submodule includes:

a clustering submodule, configured to perform clustering on the historical evaluation text to obtain a plurality of clusters corresponding to the historical evaluation text;

A splicing sub-module, for splicing historical evaluation texts whose text length is less than a preset length threshold in the cluster for each of the clusters, to obtain at least one spliced text;

The fourth determination sub-module is used to determine the subject word corresponding to the target user based on the spliced text in each cluster, the historical evaluation text whose text length is not less than the length threshold, and the subject generation model, and based on the The vector corresponding to the subject word is described, and the user vector is determined.

Optionally, the second determination submodule includes:

The fifth determination submodule is used to determine the word frequency and reverse document frequency corresponding to each word segment in the multiple evaluation texts corresponding to the target content, and the text length ratio corresponding to the word segment, wherein the text corresponding to the word segment The length ratio is the ratio of the length of the evaluation text to which the segmented word belongs and the average text length of the plurality of evaluation texts;

The sixth determination submodule is used to determine, for each of the word segmentations, the product of the word frequency corresponding to the word segmentation, the reverse document frequency and the text length ratio corresponding to the word segmentation as the target parameter corresponding to the word segmentation;

The seventh determination sub-module is configured to determine the target word segment corresponding to the target content according to the target parameter corresponding to each word segment, and determine the content vector based on the vector corresponding to the target word segment.

Optionally, the first determining module includes:

an eighth determination submodule, configured to determine the classification corresponding to the evaluation text according to the evaluation text and the text classification model, and determine the score indicated by the classification as the second evaluation result;

Wherein, in the training process of the text classification model, feature extraction is performed based on the feature extraction sub-model and target features are obtained based on the fully connected layer, and the prediction result of the text classification model is obtained by predicting some features in the target features .

Optionally, the third determining module includes:

Ninth determination submodule, for determining the weighted sum of the first evaluation result and the second evaluation result as the initial evaluation result;

An adjustment sub-module, configured to adjust the initial evaluation result according to the matching degree to obtain the quality evaluation result.

Optionally, the device further includes:

a fourth determination module, configured to determine the recommended content corresponding to the target user according to the quality evaluation result corresponding to each content in the content library;

The output module is used for outputting the recommended content.

Regarding the apparatus in the above-mentioned embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be described in detail here.

FIG. 5 is a block diagram of an electronic device 700 according to an exemplary embodiment. As shown in FIG. 5 , the electronic device 700 may include: a processor 701 and a memory 702 . The electronic device 700 may also include one or more of a multimedia component 703 , an input/output (I/O) interface 704 , and a communication component 705 .

The processor 701 is configured to control the overall operation of the electronic device 700 to complete all or part of the steps in the above-mentioned content quality evaluation method. The memory 702 is used to store various types of data to support operations on the electronic device 700, such data may include, for example, instructions for any application or method operating on the electronic device 700, and application-related data, Such as contact data, messages sent and received, pictures, audio, video, and so on. The memory 702 can be implemented by any type of volatile or non-volatile storage device or their combination, such as static random access memory (Static Random Access Memory, SRAM for short), electrically erasable programmable read-only memory ( Electrically Erasable Programmable Read-Only Memory (EEPROM for short), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), read-only Memory (Read-Only Memory, ROM for short), magnetic memory, flash memory, magnetic disk or optical disk. Multimedia components 703 may include screen and audio components. Wherein the screen can be, for example, a touch screen, and the audio component is used for outputting and/or inputting audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal may be further stored in memory 702 or transmitted through communication component 705 . The audio assembly also includes at least one speaker for outputting audio signals. The I/O interface 704 provides an interface between the processor 701 and other interface modules, and the above-mentioned other interface modules may be a keyboard, a mouse, a button, and the like. These buttons can be virtual buttons or physical buttons. The communication component 705 is used for wired or wireless communication between the electronic device 700 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, 4G, NB-IOT, eMTC, or other 5G, etc., or one or a combination of them , which is not limited here. Therefore, the corresponding communication component 705 may include: Wi-Fi module, Bluetooth module, NFC module and so on.

In an exemplary embodiment, the electronic device 700 may be implemented by one or more Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing (Digital Signal) Processing Device (DSPD for short), Programmable Logic Device (PLD for short), Field Programmable Gate Array (FPGA for short), controller, microcontroller, microprocessor or other electronic components , which is used to execute the above-mentioned content quality evaluation method.

In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, the program instructions implement the steps of the above-mentioned content quality evaluation method when executed by a processor. For example, the computer-readable storage medium can be the above-mentioned memory 702 including program instructions, and the above-mentioned program instructions can be executed by the processor 701 of the electronic device 700 to complete the above-mentioned content quality evaluation method.

FIG. 6 is a block diagram of an electronic device 1900 according to an exemplary embodiment. For example, the electronic device 1900 may be provided as a server. 6 , the electronic device 1900 includes a processor 1922 , which may be one or more in number, and a memory 1932 for storing computer programs executable by the processor 1922 . A computer program stored in memory 1932 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processor 1922 may be configured to execute the computer program to perform the content quality evaluation method described above.

In addition, the electronic device 1900 may also include a power supply assembly 1926, which may be configured to perform power management of the electronic device 1900, and a communication component 1950, which may be configured to enable communication of the electronic device 1900, eg, wired or wireless communication. Additionally, the electronic device 1900 may also include an input/output (I/O) interface 1958 . Electronic device 1900 may operate based on an operating system stored in memory 1932, such as Windows Server ^™ , Mac OS X ^™ , Unix ^™ , Linux ^™ , and the like.

In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, the program instructions implement the steps of the above-mentioned content quality evaluation method when executed by a processor. For example, the non-transitory computer-readable storage medium can be the above-mentioned memory 1932 including program instructions, and the above-mentioned program instructions can be executed by the processor 1922 of the electronic device 1900 to complete the above-mentioned content quality evaluation method.

In another exemplary embodiment, there is also provided a computer program product comprising a computer program executable by a programmable apparatus, the computer program having, when executed by the programmable apparatus, for performing the above The code section of the content quality evaluation method.

The preferred embodiments of the present disclosure have been described above in detail with reference to the accompanying drawings. However, the present disclosure is not limited to the specific details of the above-mentioned embodiments. Various simple modifications can be made to the technical solutions of the present disclosure within the scope of the technical concept of the present disclosure. These simple modifications all fall within the protection scope of the present disclosure.

In addition, it should be noted that each specific technical feature described in the above-mentioned specific implementation manner may be combined in any suitable manner under the circumstance that there is no contradiction. In order to avoid unnecessary repetition, various possible combinations are not described in the present disclosure.

In addition, the various embodiments of the present disclosure can also be arbitrarily combined, as long as they do not violate the spirit of the present disclosure, they should also be regarded as the contents disclosed in the present disclosure.

Claims (10)

1. A content quality evaluation method, characterized by comprising:

acquiring a first evaluation result and an evaluation text of a target user aiming at target content;

determining a second evaluation result of the target content according to the evaluation text;

determining the matching degree corresponding to the target user and the target content;

and determining the quality evaluation result of the target user for the target content according to the first evaluation result, the second evaluation result and the matching degree.

2. The method of claim 1, wherein determining the degree of match between the target user and the target content comprises:

determining a user vector corresponding to the target user based on the historical evaluation text of the target user;

determining a content vector corresponding to the target content based on a plurality of evaluation texts corresponding to the target content;

and determining the matching degree according to the user vector and the content vector.

3. The method of claim 2, wherein determining the user vector corresponding to the target user based on the historical rating text of the target user comprises:

clustering the historical evaluation texts to obtain a plurality of cluster clusters corresponding to the historical evaluation texts;

for each cluster, splicing the historical evaluation texts with the text length smaller than a preset length threshold value in the cluster to obtain at least one spliced text;

and determining a subject word corresponding to the target user based on the spliced text in each cluster, the history evaluation text with the text length not less than the length threshold value and a subject generation model, and determining the user vector based on the vector corresponding to the subject word.

4. The method of claim 2, wherein determining the content vector corresponding to the target content based on the plurality of rating texts corresponding to the target content comprises:

determining a word frequency and a reverse document frequency corresponding to each word in a plurality of evaluation texts corresponding to the target content, and a text length proportion corresponding to the word, wherein the text length proportion corresponding to the word is the ratio of the length of the evaluation text to which the word belongs to the average text length of the plurality of evaluation texts;

for each word segmentation, determining the product of the word frequency corresponding to the word segmentation, the reverse document frequency and the text length proportion corresponding to the word segmentation as a target parameter corresponding to the word segmentation;

and determining a target word segmentation corresponding to the target content according to the target parameter corresponding to each word segmentation, and determining the content vector based on the vector corresponding to the target word segmentation.

5. The method of claim 1, wherein determining a second rating result of the target content according to the rating text comprises:

determining a classification corresponding to the evaluation text according to the evaluation text and the text classification model, and determining the grade indicated by the classification as the second evaluation result;

in the training process of the text classification model, feature extraction is carried out based on the feature extraction submodel, target features are obtained based on the full connection layer, and the prediction result of the text classification model is obtained by predicting partial features in the target features.

6. The method according to claim 1, wherein the determining the quality evaluation result of the target user for the target content according to the first evaluation result, the second evaluation result and the matching degree comprises:

determining a weighted sum of the first and second evaluation results as an initial evaluation result;

and adjusting the initial evaluation result according to the matching degree to obtain the quality evaluation result.

7. The method according to any one of claims 1-6, further comprising:

determining recommended content corresponding to the target user according to a quality evaluation result corresponding to each content in a content library;

and outputting the recommended content.

8. A content quality evaluation apparatus, characterized in that the apparatus comprises:

the acquisition module is used for acquiring a first evaluation result and an evaluation text of a target user aiming at target content;

the first determining module is used for determining a second evaluation result of the target content according to the evaluation text;

the second determining module is used for determining the matching degree corresponding to the target user and the target content;

and the third determining module is used for determining the quality evaluation result of the target user aiming at the target content according to the first evaluation result, the second evaluation result and the matching degree.

9. A non-transitory computer readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 7.

10. An electronic device, comprising:

a memory having a computer program stored thereon;

a processor for executing the computer program in the memory to carry out the steps of the method of any one of claims 1 to 7.

Images