Brand value evaluation method and device, equipment, storage medium and program product

Abstract

The embodiment of the invention provides a brand value evaluation method, device and equipment, a computer readable medium and a computer program product. The method described herein includes: obtaining brand behavior data associated with a target brand and user evaluation data for the target brand; determining a user evaluation index for the target brand based on the user evaluation data; and determining a value attribute value of the target brand based on the brand behavior data of the target brand and the user evaluation index by using a trained value evaluation model, the value evaluation model being trained to represent an association relationship between the brand behavior data of the brand and the value attribute value of the brand and between the user evaluation index and the value attribute value of the brand. Therefore, the brand value can be automatically evaluated, and the value evaluation accuracy is improved.

Description

Technical Field

[0001] Example embodiments of the present disclosure generally relate to the field of computers, and more particularly, to methods, apparatuses, devices, computer-readable storage media, and computer program products for brand value assessment. Background Art

[0002] A brand is an inherent attribute of a good or service, used to distinguish different products or services. Nowadays, with the rapid development of e-commerce, the number of brands on various platforms is huge. How to objectively measure brand value in a systematic and intelligent manner is of great significance to the refined operation of the platform. Summary of the Invention

[0003] In a first aspect of the present disclosure, a method for brand value assessment is provided. The method includes: obtaining brand behavior data associated with a target brand and user evaluation data for the target brand; determining user evaluation indicators for the target brand based on the user evaluation data; and using a trained value evaluation model to determine a value attribute value of the target brand based on the brand behavior data and user evaluation indicators of the target brand, where the value evaluation model is trained to represent the association relationship between the brand behavior data and user evaluation indicators of the brand and the value attribute value of the brand.

[0004] In a second aspect of the present disclosure, an apparatus for brand value assessment is provided. The apparatus includes: an obtaining module configured to obtain brand behavior data associated with a target brand and user evaluation data for the target brand; a user evaluation indicator determination module configured to determine user evaluation indicators for the target brand based on the user evaluation data; and a value attribute value determination module configured to use a trained value evaluation model to determine a value attribute value of the target brand based on the brand behavior data and user evaluation indicators of the target brand, where the value evaluation model is trained to represent the association relationship between the brand behavior data and user evaluation indicators of the brand and the value attribute value of the brand.

[0005] In a third aspect of the present disclosure, an electronic device is provided. The electronic device includes at least one processing unit; and at least one memory, where at least one memory is coupled to at least one processing unit and stores instructions for execution by at least one processing unit, and when the instructions are executed by at least one processing unit, the electronic device executes the method of the first aspect of the present disclosure.

[0006] In a fourth aspect of the present disclosure, a computer-readable storage medium is provided. A computer program is stored on the computer-readable storage medium, and it can be executed by a processor to execute the method according to the first aspect of the present disclosure.

[0007] In a fifth aspect of the present disclosure, there is provided a computer program product. The computer program product includes computer-executable instructions that, when executed by a processor, implement the method according to the first aspect of the present disclosure.

[0008] It should be understood that the content described in this part is not intended to define the key features or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In the following, with reference to the accompanying drawings and the following detailed description, the above and other features, advantages, and aspects of the various implementations of the present disclosure will become more apparent. In the drawings, the same or similar reference numerals denote the same or similar elements, where:

[0010] Figure 1 A schematic diagram of an exemplary environment in which embodiments of the present disclosure can be implemented is shown;

[0011] Figure 2 A block diagram of a brand value evaluation system according to some embodiments of the present disclosure is shown;

[0012] Figure 3 A block diagram of a process of a brand value evaluation method according to some embodiments of the present disclosure is shown;

[0013] Figure 4 A block diagram of a process of discriminating an emotional attribute value using an emotional attribute discrimination model according to some embodiments of the present disclosure is shown;

[0014] Figure 5 A block diagram of a text extractor according to some embodiments of the present disclosure is shown;

[0015] Figure 6 A block diagram of a value evaluation model according to some embodiments of the present disclosure is shown;

[0016] Figure 7 A block diagram of a training process of a value evaluation model according to some embodiments of the present disclosure is shown;

[0017] Figure 8 A flowchart of a brand value evaluation process according to some embodiments of the present disclosure is shown;

[0018] Figure 9 A block diagram of a brand value evaluation device according to some embodiments of the present disclosure is shown; and

[0019] Figure 10 A block diagram of an electronic device in which one or more embodiments of the present disclosure can be implemented is shown. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although some embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. On the contrary, these embodiments are provided to more thoroughly and completely understand the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not used to limit the protection scope of the present disclosure.

[0021] In the description of the embodiments of the present disclosure, the term "including" and its similar terms should be understood as open inclusion, that is, "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The term "some embodiments" should be understood as "at least some embodiments". There may also be other explicit and implicit definitions hereinafter.

[0022] It should be noted that in the technical solution of the present disclosure, the acquisition, storage, application, etc. of the user's personal information all comply with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0023] As used herein, the term "model" can learn the association between the corresponding input and output from the training data, so that after training, for a given input, the corresponding output can be generated. The generation of the model can be based on machine learning techniques. Deep learning is a machine learning algorithm that processes the input and provides the corresponding output by using multiple processing units. In this article, the "model" can also be referred to as a "machine learning model", a "machine learning network", or a "network", and these terms can be used interchangeably in this article.

[0024] As described above, objectively measuring the brand value in a systematic and intelligent manner is of great significance for the refined operation of various platforms (especially e-commerce platforms).

[0025] In some solutions, when measuring the brand value, relatively single indicators are used. For example, behavioral indicators of the brand on the platform such as sales volume or search volume set rule thresholds or indicator weightings from one aspect or several aspects, so as to divide the levels of brands within the platform.

[0026] However, this method is not objective and comprehensive enough for measuring the brand value. Using relatively single indicators or certain indicators to set rules or weightings for certain aspects of the brand only represents the performance of the brand included in the indicators in the platform, and the selection of brand behavioral indicators and the setting of rules have strong subjective factors.

[0027] In some other solutions, relying on the impressions and understanding of brands by humans, the value of some brands in a certain category is evaluated. Different personnel divide the brands they are responsible for according to different scenarios in a targeted manner.

[0028] However, this manual discrimination method will consume a large amount of labor costs and cannot measure the value of all brands. Moreover, the method of researching brands by humans or judging brand value based on experience requires a relatively high level of industry knowledge background for personnel. It is quite difficult to gather industry professionals for all the product brands on the platform. Even if industry professionals from various industries are gathered, it cannot be guaranteed that they have a clear understanding of all brands in their respective industries. Therefore, the number of brands that this method can cover is small and cannot cover all brands on the platform.

[0029] In summary, the current methods for evaluating brand value are relatively single and subjective, often being manual discrimination within a small range of each industry, lacking a unified measurement method. Moreover, the measurement of brand value and the application after brand value measurement do not form a complete chain. For example, when there are quantity restrictions on the brands that can participate in certain activities on the platform, usually humans screen and report brands based on experience. For example, marketing personnel in each industry manually screen a small number of brands that meet a certain requirement based on experience and submit and approve them step by step for application.

[0030] Example Environment and Basic Working Principle

[0031] In the conventional solution, the method for measuring brand value is relatively single and subjective, lacking an automated tool to intelligently evaluate brand value. This solution proposes a brand value evaluation method based on a value evaluation model. According to the embodiments of the present disclosure, for a target brand that needs to be evaluated for value, brand behavior data associated with the target brand and user evaluation data for the target brand are obtained. Based on the user evaluation data, user evaluation indicators for the target brand are determined. Then, the brand behavior data and the user evaluation indicators are used as inputs to the value evaluation model, and the value attribute value of the target brand is obtained using the value evaluation model as the value evaluation result of the target brand, thereby realizing the evaluation of the value of the target brand.

[0032] The solution provided by the embodiments of the present disclosure obtains brand behavior data and user evaluation data associated with a target brand, and constructs a value evaluation model based on a machine learning algorithm. The value evaluation model is used to evaluate the value of the target brand. Compared with single brand behavior data, the brand behavior data obtained in the present disclosure can be various types of brand behavior data, and user evaluation indicators are taken into consideration. Therefore, the value evaluation model considers more comprehensive indicators, and the brand value evaluation result obtained by this method is more objective and accurate. Moreover, this method can be applied to any brand, with a wide range of applications, which helps to build the brand ecosystem of the platform. In addition, this method does not require manual or industry experts to evaluate the brand value of all brands, so it can reduce labor costs.

[0033] Example embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0034] Figure 1 A schematic diagram of an example environment 100 in which the embodiments of the present disclosure can be implemented is shown. Environment 100 involves a model application device 120, which is configured to evaluate the value of a target brand. For example, the target brand can be a brand in any industry and any category. A trained value evaluation model 130 is deployed in the model application device 120. In the embodiments of the present disclosure, the value evaluation of the target brand is performed by means of the trained value evaluation model 130.

[0035] The value evaluation model 130 can be configured as any suitable type of model. In some embodiments, the value evaluation model 130 can be based on the DCN (Deep&Cross Network) model. In some embodiments, the value evaluation model 130 can also be based on classification models such as GBDT (Gradient Boosting Decision Tree), DeepFM (deep factor machine), etc. Of course, any other similar models are also applicable.

[0036] The model application device 120 can utilize the value evaluation model 130 deployed thereon to evaluate the value of each brand (for the sake of discussion herein, the brand to be evaluated is referred to as the target brand). When using the value evaluation model 130 for evaluation, the brand behavior data 110 and the user evaluation data 112 associated with the target brand are considered. The brand behavior data 110 can be input into the value evaluation model 130 as the characteristic data of the target brand. For the user evaluation data 112, the user evaluation indicators 115 for the target brand can be determined from the user evaluation data 112 and input into the value evaluation model 130 as the characteristic data of the target brand. The user evaluation data 112 can include user comments 111 for the target brand. The user evaluation indicators 115 for the target brand can be obtained based on the user comments 111 for the target brand. Further, the model application device 120 can utilize the value evaluation model 130 to generate a value attribute value 140 to indicate the value evaluation result of the target brand.

[0037] In the environment 100, the model training device 125 is configured to train the value evaluation model 130. The training of the value evaluation model 130 can be implemented by any suitable computing system or device, such as a server, a cloud computing device, an edge computing node, etc. In some embodiments, the training of the value evaluation model 130 can be completed at the model training device 125, and the trained value evaluation model 130 can be provided to the model application device 120 for use. In some embodiments, the training of the value evaluation model 130 can be partially or fully implemented at the model application device 120.

[0038] In the environment 100, the model application device 120 or the model training device 125 can be any type of device with computing capabilities, including a terminal device or a server device. The terminal device can be any type of mobile terminal, fixed terminal or portable terminal, including a mobile phone, a desktop computer, a laptop computer, a notebook computer, a netbook computer, a tablet computer, a media computer, a multimedia tablet, a personal communication system (PCS) device, a personal navigation device, a personal digital assistant (PDA), an audio / video player, a digital camera / video camera, a positioning device, a television receiver, a radio broadcast receiver, an e-book device, a game device, or any combination of the foregoing, including accessories and peripherals of these devices or any combination thereof. The server device can, for example, include a computing system / server, such as a mainframe, an edge computing node, an electronic device in a cloud environment, etc. Although shown as separate devices, the model application device 120 and the model training device 125 can be the same device, or be included in the same system.

[0039] It should be understood that the structure and functions of the environment 100 are described for exemplary purposes only, without implying any limitation on the scope of the present disclosure.

[0040] Next, for a clearer understanding of the present disclosure, taking the evaluation of the brand value of a platform as an example, an exemplary structure of a brand value evaluation system for implementing the brand value evaluation method of the present disclosure is first introduced.

[0041] Example Embodiment of Brand Value Evaluation System

[0042] Figure 2 A system 200 for brand value evaluation according to some embodiments of the present disclosure is shown, which can implement processes such as value measurement of brands in a platform, application of brand value determination, and feedback update between different subsystems / modules.

[0043] In the embodiments of the present disclosure, brand value measurement can also be described as brand value evaluation, brand value determination, etc. The value evaluation model can also be described as a value measurer, etc.

[0044] It should be noted that the present disclosure is not limited to the modules / systems involved in the brand value evaluation system 200. The brand value evaluation system 200 may include more or fewer modules, and adding or replacing modules / systems in the figure to achieve a similar purpose should be regarded as being included in the scope of the present disclosure.

[0045] In addition, the above brand value evaluation system includes both modules involved in the training process of the value evaluation model and modules involved in the use process of the value evaluation model. Generally, the training of the value evaluation model and the use of the value evaluation model are executed on different devices and systems. For example, Figure 2 the sample extraction module 210, label formulation module 215, data acquisition and feature engineering module 220, user evaluation index construction module 230, and the training process of the value evaluation model 130 in Figure 1 can be executed on a separate system, such as Figure 1 implemented in the model training device 125 in

[0046] The design and implementation of the brand value evaluation system are described below from two aspects: core algorithm capabilities and the interaction logic between modules / subsystems within the system.

[0047] First, the core algorithm capabilities part is introduced.

[0048] Core algorithm capabilities ( Figure 2which is shown by a dashed box) includes two major parts: model data preparation and brand value evaluation.

[0049] Referring to Figure 2 , the modules for implementing model data preparation include a sample extraction module 210, a label formulation module 215, and a data acquisition and feature engineering module 220.

[0050] Among them, the sample extraction module 210 and the label formulation module 215 are interrelated. In some embodiments of the present disclosure, in the construction of the initial sample set (which can also be described as the initial training set), based on the platform's consideration of the brand ecosystem and experts' understanding of the brand status in various industries, through means such as clustering and mining the probability distribution information of examples, the sample extraction and label formulation of the benchmark classifier are reasonably planned. Specifically, the brand ecosystem is determined by the perception of the brand by platform business personnel or the market. As an example, the value attribute value of a brand can be defined as the following three brand levels: The first level is the top brand, which has high popularity, high brand credibility, quality, and association power, and has a market leadership position. The second level is the well-known brand, which has a specific culture and personality, has formed stable brand preferences and brand credibility, and a stable customer group. The third level is the general brand, which has a certain popularity or only a trademark and can meet the functional needs of consumers. For different industries, the proportion of brands at different levels will vary. For example, in the fashion industry, the proportion of brands at the first level and the second level, that is, the top brands and well-known brands, may be more in the overall industry than in related industries such as fresh food. Then, when extracting samples and formulating labels, the overall principle is to make the distribution of sample brands at all levels in all industries in the selected sample brands as consistent as possible with the overall distribution of brands at all levels in all industries. Random extraction is carried out on the basis of preliminary clustering and according to the number of various types of brands recommended by experts in each industry. Further, labels are formulated based on experts' value judgments of the brand. For example, three types of labels: high value, medium value, and low value. Another example is that, similar to the brand level, three types of labels: the first level, the second level, and the third level are formulated. In the embodiments of the present disclosure, different labels are intended to distinguish different values of the brand, and the present disclosure does not limit the label formulation method. In addition to discrete classification of the value attribute value of the brand, in some embodiments, the value attribute value of the brand can also be defined as a continuous value, for example, in the value range from 0 to 1, where the larger the value, the higher the brand value.

[0051] The data acquisition and feature engineering module 220 is configured to combine the characteristics of the brand, fully consider various data sources inside and outside the platform, and obtain feature data by considering the information of the brand in its respective industries from the perspective of data modalities.

[0052] The feature data obtained by the data acquisition and feature engineering module 220 includes various types of brand behavior data. For example, the brand behavior data includes channel information associated with the brand, such as the total number of offline stores corresponding to the brand and their respective proportions, the number of platforms entered, etc. Volume information associated with the brand, such as scale and market share, transaction amounts on various platforms, etc. Popularity information associated with the brand, such as whether there is a spokesperson, the comprehensive index of website spokespersons, discussion degree, etc. Network metrics associated with the brand, such as the number of followers of each platform account, search exposure click-through numbers, positive review rates, negative review rates, proportions of various types of users, etc. And information related to the user group associated with the brand and the portrait of the people who have purchased the brand.

[0053] In addition, a brand is an overall image expression of an enterprise. In the current era of rapid network development, it inevitably interacts with the main viewpoints of public opinions on brand-related events, that is, public opinions. The brand value will also be reflected or affected to a certain extent. Therefore, it is very necessary to introduce as much effective information in brand public opinions into brand value measurement as possible.

[0054] Therefore, in some embodiments of the present disclosure, the data obtained by the data acquisition and feature engineering module 220 further includes user comments on the brand. Among them, the user comments on the brand include the remarks about the brand made by users on various platforms in the network, etc.

[0055] The user comments on the brand can include multiple user comments made by different users on different network platforms. Analyzing the user comments can determine the degree of users' liking for the brand. Therefore, in some embodiments of the present disclosure, the user evaluation index construction module 230 can construct user evaluation indexes based on the user comments on the brand obtained by the data acquisition and feature engineering module 220.

[0056] In some embodiments of the present disclosure, the user comments on the brand can also be described as data related to brand public opinions or brand public opinion data. The user evaluation index can also be described as a public opinion index, which can represent the degree of users' liking for the brand.

[0057] In some embodiments of the present disclosure, based on the user comments on the brand, the emotional polarity of the user comments on the brand is discriminated to obtain the emotional attribute value corresponding to each user comment, and each emotional attribute value can represent a type of emotion. Then, based on the weighted calculation of the proportion of the number of different emotional categories corresponding to the discriminated emotional attribute values, a user evaluation index is constructed.

[0058] After that, based on the brand behavior data obtained by the data acquisition and feature engineering module 220 and the constructed user evaluation indicators, a value evaluation model 130 is trained. The trained value evaluation model 130 can represent the association relationship between the brand behavior data and user evaluation indicators of the brand and the value attribute value of the brand. In addition, as described above, the value evaluation model 130 can be implemented as a DCN model, and can also be implemented as models such as GBDT and DeepFM. The embodiments of the present disclosure do not limit the specific implementation manner of training the value evaluation model. The trained value evaluation model can be retained in the big data platform 260, and further, can be stored in the database 270.

[0059] The following describes the process of determining the brand value by applying the trained value evaluation model 130.

[0060] Refer to Figure 3 , for the target brand whose brand value is to be determined, the process 300 of determining the value of the target brand includes: obtaining the characteristic data of the target brand, including user comments 111 for the target brand and brand behavior data 110 associated with the target brand. The brand behavior data associated with the target brand includes at least one of the following: channel information associated with the target brand, volume information associated with the target brand, popularity information associated with the target brand, network metrics associated with the target brand, and user groups associated with the target brand. The specific implementation of these behavior data can refer to the foregoing description and will not be elaborated here.

[0061] Furthermore, for the user comments 111 for the target brand, the sentiment attribute discrimination model 310 is used to obtain the sentiment attribute value 320 included in the user comments, which is used to represent the sentiment polarity in the user comments. And based on the multiple sentiment attribute values corresponding to multiple comments, a user evaluation indicator 115 for the target brand is obtained.

[0062] In the embodiments of the present disclosure, for a given comment, the sentiment discrimination model can combine the context information included in the comment to determine the sentiment polarity included in the given comment. The sentiment discrimination model can be implemented as two parts: a text extractor + a context information extraction model.

[0063] As an example, refer to Figure 4 , the sentiment discrimination model 310 includes a text extractor 420 and an attention-based model 440. The attention-based model 440 can be used for context information extraction.

[0064] For a given comment, its processing procedure 400 includes using a text extractor 420 to extract a text feature vector 430 from the given comment text 410. Further, using an attention-based model 440, the attention weights 450 of multiple text units in the given comment are determined from the text feature vector 430, and the attention weights can express the context information in the given comment. Based on the text feature vector 430 and the determined attention weights 450, the sentiment attribute value corresponding to the given comment is obtained through a fully connected classification layer 460 (e.g., a classification layer implemented by softmax).

[0065] In some embodiments, in the process of discriminating the sentiment polarity of user comments about a brand, the idea of the RoBERTa (A Robustly Optimized BERT) model can be borrowed and the context information can be fully considered. Therefore, as an example, a text classification model of RoBERTa-BiGRU-Att can generally be adopted. For example, the text extractor 420 can be configured as a RoBERTa model. The model for extracting context information can be configured as an attention-based model, such as being configured as BiGRU-Att (Bidirectional Gated Recurrent Unit-Attention). BiGRU-Att includes a BiGRU model and an Attention layer. BiGRU superimposes the feature information through the gru in the forward and backward passes, so that the feature information is retained to the greatest extent, and the trained feature vector is then input into the attention layer. And, the embedding layer vector representation of the RoBERTa model is input into BiGRU-Att to obtain comprehensive context information, and then classified through a fully connected layer. And, a cross-entropy loss function can be adopted.

[0066] Among them, RoBERTa, that is, simply and roughly called the strongly optimized BERT method, is a language representation model. The RoBERTa model is evolved from the BERT model, and its structure is the same as that of the BERT model. The following improvements are made in the training method: deleting the next sentence prediction (NSP) task. Adjusting the static mask to a dynamic mask, generating a mask dynamically once before each training, so that the model can learn more sentence patterns. And, using byte pair encoding to process text data. In addition, the RoBERTa model is strengthened in terms of the scale, computing power, and data of the BERT model.

[0067] GRU (Gated Recurrent Unit) can capture long-distance semantic information. BiGRU, that is, a bidirectional gated recurrent unit structure, solves the long-order dependence problem and considers both directions at the same time. That is, it processes the information in both the front and back directions of the text, can consider the context of the user comment, and can represent the text information more fully.

[0068] As an example, Figure 5 an example structure of the RoBERTa model is shown.

[0069] As Figure 5 shown, the initial input is a piece of text, which can be represented by S 1 , S 2 , …, S n (as indicated by label 510). The sum of the text word vectors, sentence vectors, and position vectors is used as the input to the RoBERTa model, which is represented by E 1 , E 2 , …, E n (as indicated by label 520). The intermediate layer represents a multi-layer bidirectional feature extractor (Transformer) 530, and the output vector 540 of the model can be represented by T 1 , T 2 , …, T n .

[0070] In addition, Figure 5 an architecture 535 of the feature extractor 530 is also shown in , and this architecture 535 includes processing procedures such as a multi-head attention mechanism, residual connection & layer normalization, a feed-forward neural network, and residual connection & layer normalization.

[0071] Alternatively or additionally, embodiments of the present disclosure can also modify the language of the RoBERTa model. For example, change its language from English to Chinese to make it adapt to relevant marketing words or hot words in the brand, etc.

[0072] The above takes RoBERTa - BiGRU - Att as an example for illustration. In other embodiments, the text extractor 420 can also be configured as other models suitable for extracting text features, such as the BERT model. Similarly, the model for determining context information can also be configured as other models, such as the LSTM model. Embodiments of the present disclosure are not limited in this regard.

[0073] Through the above process, corresponding sentiment attribute values can be obtained for each of the multiple comments, that is, multiple sentiment attribute values.

[0074] Further, the multiple sentiment attribute values 320 are processed to obtain a user evaluation index 115 for the target brand. The following details this processing process.

[0075] In some embodiments of the present disclosure, the emotional attribute value indicates one emotional category among multiple emotional categories. And determining the user evaluation index based on multiple emotional attribute values includes: respectively determining the proportion of the emotional attribute values indicating the multiple emotional categories among the multiple emotional attribute values; and using the weights of the multiple emotional categories respectively, to determine the user evaluation index by weighting the proportions of the emotional attribute values of the multiple emotional categories.

[0076] That is, for different emotional attribute values, a weighted calculation is performed on the proportion of the quantity in all emotional attribute values, to construct a user evaluation index for the target brand.

[0077] Exemplarily, the emotional categories corresponding to the emotional attribute values 0, 1, and 2 are three emotional categories: good, medium, and bad respectively. Then, among the emotional attribute values corresponding to multiple comments, the ratios of the emotional attribute values with values of 0, 1, and 2 respectively in the total quantity of all emotional attribute values are respectively counted. For example, the total quantity of all emotional attribute values is 100. Among them, the total number of emotional attribute values of 0 is 50, that is, its proportion is 50%, indicating that the proportion corresponding to the emotional category good is 50%. The numbers of emotional attribute values 1 and 2 are 30 and 20 respectively, that is, their proportions are 30% and 20% respectively, indicating that the proportions corresponding to the emotional categories medium and bad are 30% and 20% respectively. In addition, the weights of the three emotional categories of good, medium, and bad are 0.4, 0.3, and 0.3 respectively. Then, the user evaluation index of the target brand can be obtained as: 0.5 * 0.4 + 0.3 * 0.3 + 0.2 * 0.3 = 0.35. Suppose that when the value of the user evaluation index is in the range of [0, 0.4), it indicates that the user's degree of liking for the brand is poor. When the value is in the range of [0.4, 0.7), it indicates that the user's degree of liking for the brand is at a medium level. When the value is in the range of [0.7, 1), it indicates that the user's degree of liking for the brand is high. Then, the value of the above user index being 0.35 indicates that the user's degree of liking for the brand is poor.

[0078] After that, based on the constructed user evaluation index 115 for the target brand, and the obtained behavioral data 110 associated with the target brand, the value of the target brand is evaluated using the value evaluation model 130.

[0079] The DCN model is a deep model that can efficiently learn low-dimensional feature crosses and high-dimensional non-linear features simultaneously. In some embodiments of the present disclosure, based on the idea of the DCN algorithm, the value evaluation model 130 is designed. In this way, the value evaluation model 130 can automatically learn the high-order interaction effects between various features.

[0080] Continue to refer to Figure 3, in some embodiments of the present disclosure, the value evaluation model 130 includes at least an interaction network 340 and a deep network 350. Correspondingly, determining the value attribute value of the target brand includes: converting the brand behavior data 110 and the user evaluation metrics 115 into a combined feature vector 330.

[0081] In some embodiments of the present disclosure, the brand behavior data includes multiple types of brand behavior data; and converting the brand behavior data and the user evaluation metrics into a combined feature vector includes: converting the multiple types of brand behavior data into multiple behavior feature vectors respectively; classifying the user evaluation metrics and the multiple behavior feature vectors into a continuous feature set and a discrete feature set, and the user evaluation metrics are classified into the continuous feature set. Then, converting the feature vectors in the discrete feature set into feature vectors represented by dense vectors; and combining the continuous feature set and the feature vectors represented by dense vectors into a combined feature vector.

[0082] Specifically, there are numerous behavior characteristics of a brand, which characterize the value of the brand from many aspects. These characteristics are both continuous and discrete, rich and heterogeneous in content, and display different information from different dimensions. Multiple brand behavior data can be converted into multiple behavior feature vectors, and multiple feature vectors and user evaluation metrics can be classified. For example, the user evaluation metrics can be classified as continuous features. The channel information associated with the target brand, such as the total number of offline stores and their respective proportions, and the number of platforms entered, is classified as continuous features. The volume information associated with the target brand, such as the transaction amount on each platform and the behavior feature vectors after conversion of the market share, is classified as continuous features. The network metrics associated with the target brand, such as the number of followers on each platform account, the number of search exposures, clicks, the positive review rate, the negative review rate, and the behavior feature vectors after conversion of the proportion of each type of user, are classified as continuous features. The popularity information associated with the target brand, such as the behavior feature vectors after conversion of whether there is a spokesperson, is classified as discrete features, and the comprehensive index of the website spokesperson and the behavior feature vectors after conversion of the degree of discussion are classified as continuous features. The behavior feature vectors after conversion of the user portrait information of the user group associated with the target brand are also classified. For example, for the purchasing users of the target brand, since there are many users of the target brand, the user portrait information can be divided into continuous features: such as the number of male users, the number of female users, the number of elderly users, etc.; or the target users of the target brand can be normalized. For example, for a brand related to women's products, the user portrait information is single at this time, and it is a discrete feature at this time. In this way, a discrete feature set and a continuous feature set can be obtained. Then, the discrete features can be converted into dense vectors with real values and combined with the continuous features to form a combined vector.

[0083] After that, using the cross network 340, perform feature crossing on the combined feature vector representation 330 to obtain the first feature 345. Using the deep network 350, perform multiple feature extractions on the combined feature vector representation 330 to obtain the second feature 355. Moreover, determine the value attribute value 140 of the target brand based on the first feature 345 and the second feature 355.

[0084] The above value evaluation process is described below using the value evaluation model as the DCN model.

[0085] Reference Figure 6 , the model structure of DCN includes an embedding stacking layer 610, a cross network 620, a deep network 625, and a connection output layer 630.

[0086] First, in the embedding stacking layer (in Figure 6 , the embedding layer and the stacking layer are combined and represented as the embedding stacking layer) 610, decompose the discrete features and continuous features. Specifically, in the embedding layer, convert the discrete features 612 into dense vectors of real values, and form the vector x with the continuous features 614 through the stacking layer 0。 Furthermore, input the vector x 0 into the cross network and the deep network. Among them, the cross network 620 includes L 1 cross layers. The deep network includes n deep layers. Exemplarily, the deep network can be a fully connected neural network with forward propagation, using Relu as the activation function. The cross network applies explicit feature crossing. Each layer will add the crossed features and the bias term and then add the input data of this layer, and can have up to L 1 +1-order feature crossing, with high-dimensional feature crossing ability. Exemplarily, the processing of the first deep layer on the vector X 0 can be h 1 = Relu(W h，0 x 0 +b h，0 ). The processing of the first cross layer on the vector X 0 can be, It can be understood that subsequent cross layers and deep layers perform similar processing on the output of the previous layer. The connection output layer 630 connects the final outputs of the deep network 625 and the cross network 620, and obtains the connection result x stack after weighted summation, and then generates the final probability value by the softmax function for value discrimination. For example, perform the following processing using the softmax function: p = softmax(W logit x stack +b logit) where p represents the corresponding probability value. In the embodiments of the present disclosure, Softmax maps the input values of the last layer to real numbers between 0 and 1, and ensures through normalization that the sum of the brand attribute values (corresponding to n categories of brand values) of n categories is 1. The category of brand value corresponding to the brand attribute value with the highest probability is the brand value category determined by the value evaluation model.

[0087] In addition, the DCN model can use cross-entropy as the loss function to train the DCN model. The trained model can be retained on the big data platform 260.

[0088] In addition, considering that the same brand may be involved in multiple industries, the value evaluation results of the brand in different industries may be different. Therefore, the value attribute values of a certain brand in different industries can be determined. Then, when obtaining the user evaluation data and the behavioral data of the target brand, the selection of at least one industry related to the target brand by users can be determined first, and then the brand behavioral data and user evaluation data of the target brand in at least one industry can be obtained. And based on the obtained data, the value evaluation model 130 is used to judge the brand value of the target brand in at least one industry. Correspondingly, the value attribute value of the target brand indicates the value attribute value of the target brand in at least one industry.

[0089] Thus, the process of evaluating the brand value using the value evaluation model 130 is completed.

[0090] Furthermore, the value evaluation results of the target brand evaluated by the value evaluation model 130 can be further applied as follows.

[0091] The output of the value evaluation model 130 is the value of the target brand in each industry. The same brand may be involved in multiple industries, and the value evaluation results of the brand in different industries may be different. In application, the operation personnel will select the value of the brand in some industries or customize the single brand value for application according to their respective demands. The operation personnel view the results predicted by the value evaluation model 130 through the data dashboard and the self-service data extraction tool 250. The value determination application module 240 classifies and sorts the important information of the brand according to the business scenario, and the operation personnel select and apply the application results through the P-end operation system 280 by self-configuring rules. For example, operate the brands with different values and their related products, stores, etc. targeted, or the screening results can be used in various marketing activities and traffic fields to maximize the resource efficiency.

[0092] For example, first, the value assessment model 130 is used to predict the value of brand A. For example, the prediction result is that brand A has a higher value in industry 1, such as its value attribute value is 1, indicating that it is a top brand. Then, the operator can view the results of the above value assessment through the data dashboard and the self-service data extraction tool 250. If the current business scenario is "brands in industry 1 are promoting sales", the value judgment application module 240 is used to classify multiple brands involved in industry 1, such as classification according to value attribute values: Category 1 (highest value): top brands. Category 2: well-known brands. Category 3: general brands. After that, "brand value" can be regarded as a product selection indicator, such as selecting indicators such as brand value above well-known brands and transaction amount ranking in the top 10% under the third-level category in the past 30 days to select some brands of goods and configure specific promotion rules for these goods. For another example, the business scenario is: general category activity scenario. If you want to select good brands that do not distinguish between industries, you can customize some rules to determine the single value of the brand: for example, take the highest value among the top three industry values ​​of each brand as the overall value of the brand.

[0093] In this way, according to the solution of the embodiment of the present disclosure, a complete intelligent and systematic brand value unified evaluation and application solution is constructed. When applying the brand value evaluation results, a single-level configuration and label interactive update link is set for the brand in its industry or custom brand, and the operator configures the system according to their respective demands for brand value, which has more application flexibility.

[0094] In order to reduce the labor cost of brand value assessment, and also to correct the ability of the value assessment model to misjudge the brand value. Figure 2 In some embodiments of the present disclosure, based on the active learning concept, the core algorithm capability module ( Figure 2 The module contained in the dotted box is the module of core algorithm capabilities) and has certain logical interactions with the P-side operation system 280 and the expert evaluation system 290.

[0095] refer to Figure 7 , the interactive process 700 can be used to update the value assessment model.

[0096] In some embodiments of the present disclosure, the training of the value assessment model 130 includes two stages:

[0097] In the first training phase, the value assessment model 130 is trained using a first training data set. The first training data set includes brand behavior data and user evaluation data of a plurality of first sample brands and marked value attribute values. Figure 7 In the first training phase, the first training data set (also described as the initial labeled training set) L is used. 0The value evaluation model 130 is trained using 710. Among them, the first training dataset L 0 is generated from an ecological perspective using means such as clustering.

[0098] After that, using the value evaluation model 130 trained in the first training stage, multiple value evaluation results corresponding to the multiple second sample brands are determined based on the respective brand behavior data and user evaluation data of the multiple second sample brands. Among them, the value attribute values of the multiple second sample brands are not marked.

[0099] For example, referring to Figure 7 , for the unlabeled training set U720, by passing the multiple second sample brands in the training set U through the value evaluation model 130, multiple value evaluation results can be obtained. Among them, the number of value evaluation results can be determined according to the number of categories used for pre-classifying brands by value. For example, if divided into three categories, then there are three corresponding value evaluation results.

[0100] Furthermore, based on the multiple value evaluation results, at least one second sample brand to be labeled is selected from the multiple second sample brands.

[0101] That is, at least one second sample brand to be labeled can be selected from these multiple unlabeled second sample brands. For example, at least one second sample brand to be labeled can be selected from the unlabeled training set U720. Figure 7 in the training dataset U 1 725 represents the training set composed of the selected second sample brands to be labeled. For example, select the second sample brands with inaccurate prediction results of the value evaluation model for labeling.

[0102] In some embodiments, each value evaluation result among the multiple value evaluation results can indicate multiple probabilities that the corresponding second sample brand belongs to multiple value categories. Then, correspondingly, selecting at least one second sample brand to be labeled from the multiple second sample brands includes: for each second sample brand among the multiple second sample brands, if the difference degree between at least two of the multiple probabilities indicated by the value evaluation result corresponding to the second sample brand is lower than a preset threshold, then select the second sample brand.

[0103] For example, adopt the margin sampling method in the uncertainty sampling of the active learning query strategy to select the second sample brand with the smallest difference between the largest and the second largest probabilities predicted by the model.

[0104] For another example, for a certain unlabeled sample, if the probabilities of multiple value categories indicated by its corresponding multiple value evaluation results are all very close, for example, the probabilities of three value categories are 33%, 33%, and 34% respectively, or the difference between the largest and the second largest probabilities is very small (for example, among three categories, the probabilities of two categories are 51% and 48%), it means that the value evaluation model is unable to accurately predict the value classification of this sample brand. Therefore, this sample brand needs to be labeled for further training of the model.

[0105] Furthermore, obtain the labeled value attribute values of at least one second sample brand respectively to obtain a second labeled training data set. For example, transmit at least one second sample brand to an expert evaluation system, and the expert evaluation system evaluates to obtain the value attribute value of this second sample brand. After expert evaluation, this part of the samples becomes the labeled training set L. Therefore, this second labeled training data set includes the brand behavior data, user evaluation data, and labeled value attribute values of at least one second sample brand respectively.

[0106] Furthermore, in the second training stage, use the second labeled training data set to update the value evaluation model obtained by training in the first training stage.

[0107] That is, backpropagate the second labeled training data set L 715 to the value evaluation model 130 to update the value evaluation model 130. In addition, update the unlabeled training data set U 720, that is, remove the training data in the second labeled training data set L 715 from the unlabeled training data set U 720.

[0108] In addition, the prediction result of the value evaluation model 130 for the second sample brand can also be output to the data dashboard and self-service data extraction tool 250. The operation staff will also regularly analyze the prediction result of the value evaluation model 130 through the P-side operation system 730, and feedback the unlabeled samples with questions about the brand value evaluation result to the expert evaluation system 740. And after being evaluated by the expert evaluation system 740, this part of the samples is labeled, and this part of the samples is used to further train and update the value evaluation model. In this way, select the value evaluation results with questions and input them into the expert evaluation system, and the expert evaluation system makes a judgment. And based on the judgment result, further update the training set to form a dynamic update mechanism of the model. It can dynamically expand the samples in the training set and continuously increase the accuracy of the training set. In addition, various feature data associated with the brand (such as brand behavior data and user comments on the brand) will also be dynamically adjusted over time to improve the accuracy and generalization ability of the value evaluation model and reduce the labor cost.

[0109] Example Process

[0110] Figure 8 FIG. 800 is a flowchart showing a process for brand value assessment according to some embodiments of the present disclosure. The process 800 may be implemented at the model application device 120.

[0111] At block 810, the model application device 120 obtains brand behavior data associated with the target brand and user evaluation data for the target brand.

[0112] At block 820, the model application device 120 determines user evaluation metrics for the target brand based on the user evaluation data.

[0113] At block 830, the model application device 120 uses the trained value assessment model to determine the value attribute value of the target brand based on the brand behavior data and user evaluation metrics of the target brand. The value assessment model is trained to represent the association between the brand behavior data and user evaluation metrics of the brand and the value attribute value of the brand.

[0114] In some embodiments, the user evaluation data includes multiple comments of multiple users on the target brand, and determining user evaluation metrics for the target brand based on the user evaluation data includes: using an emotion attribute discrimination model to respectively determine multiple emotion attribute values of the multiple comments; and determining user evaluation metrics based on the multiple emotion attribute values.

[0115] In some embodiments, each of the multiple emotion attribute values indicates one of multiple emotion categories, and determining user evaluation metrics based on the multiple emotion attribute values includes: respectively determining, based on the multiple emotion attribute values, the proportion of the emotion attribute values indicating the multiple emotion categories in the multiple emotion attribute values; and using the weights of the respective multiple emotion categories to determine user evaluation metrics by weighting the proportions of the emotion attribute values of the multiple emotion categories.

[0116] In some embodiments, the emotion attribute discrimination model includes a text extractor and an attention-based model, and determining multiple emotion attribute values of the multiple comments includes: for a given comment among the multiple comments, using the text extractor to extract a text feature vector from the given comment; using the attention-based model to determine the attention weights of the respective multiple text units in the given comment from the text feature vector; and determining the emotion attribute value of the given comment based on the text feature vector and the determined attention weights.

[0117] In some embodiments, the value evaluation model includes at least a cross network and a deep network, and determining the value attribute value of the target brand includes: converting brand behavior data and user evaluation metrics into a combined feature vector; using the cross network to perform feature crossing on the combined feature vector representation to obtain a first feature; using the deep network to perform multiple feature extractions on the combined feature vector representation to obtain a second feature; and determining the value attribute value of the target brand based on the first feature and the second feature.

[0118] In some embodiments, the brand behavior data includes multiple types of brand behavior data; and converting the brand behavior data and user evaluation metrics into a combined feature vector includes: converting the multiple types of brand behavior data into multiple behavior feature vectors respectively; classifying the user evaluation metrics and the multiple behavior feature vectors into a continuous feature set and a discrete feature set, with the user evaluation metrics being assigned to the continuous feature set; converting the feature vectors in the discrete feature set into feature vectors with dense vector representations; and combining the continuous feature set and the feature vectors with dense vector representations into the combined feature vector.

[0119] In some embodiments, the training of the value evaluation model includes: in the first training stage, training the value evaluation model using a first training data set, the first training data set including the brand behavior data and user evaluation data of multiple first sample brands and the marked value attribute values; using the value evaluation model trained in the first training stage to determine multiple value evaluation results corresponding to multiple second sample brands based on the brand behavior data and user evaluation data of the multiple second sample brands, the value attribute values of the multiple second sample brands not being marked; based on the multiple value evaluation results, selecting at least one second sample brand to be marked from the multiple second sample brands; obtaining the marked value attribute values of the at least one second sample brand respectively to obtain a second marked training data set, the second marked training data set including the brand behavior data and user evaluation data of the at least one second sample brand respectively and the marked value attribute values; and in the second training stage, using the second marked training data set to update and train the value evaluation model.

[0120] In some embodiments, each of the multiple value evaluation results indicates multiple probabilities that the corresponding second sample brand belongs to multiple value categories, and selecting at least one second sample brand to be marked from the multiple second sample brands includes: for each of the multiple second sample brands, if the difference degree between at least two of the multiple probabilities indicated by the value evaluation result corresponding to the second sample brand is lower than a preset threshold, selecting the second sample brand.

[0121] In some embodiments, the brand behavior data associated with the target brand includes at least one of the following: channel information associated with the target brand, volume information associated with the target brand, popularity information associated with the target brand, network metrics associated with the target brand, and user groups associated with the target brand.

[0122] In some embodiments, obtaining brand behavior data associated with the target brand and user evaluation data for the target brand includes: determining a user selection in at least one of a plurality of industries related to the target brand; and obtaining brand behavior data and user evaluation data of the target brand within at least one industry; and wherein the value attribute value of the target brand indicates the value attribute value of the target brand within the at least one industry.

[0123] Example Device

[0124] Figure 9 A block diagram of a brand value evaluation model 900 according to some embodiments of the present disclosure is shown. The device 900 may be implemented in the model application device 120. Each module / component in the device 900 may be implemented by hardware, software, firmware, or any combination thereof.

[0125] The device 900 includes an acquisition module 910 configured to acquire brand behavior data associated with the target brand and user evaluation data for the target brand. The device 900 further includes: an evaluation index determination module 920 configured to determine user evaluation indexes for the target brand based on the user evaluation data. The device 900 further includes: a value determination module 930 configured to use a trained value evaluation model to determine the value attribute value of the target brand based on the brand behavior data and user evaluation indexes of the target brand, and the value evaluation model is trained to represent the association relationship between the brand behavior data and user evaluation indexes of the brand and the value attribute value of the brand.

[0126] In some embodiments, the user evaluation data includes multiple comments of multiple users on the target brand, and wherein determining user evaluation indexes for the target brand based on the user evaluation data includes: using an emotion attribute discrimination model to respectively determine multiple emotion attribute values of the multiple comments; and determining user evaluation indexes based on the multiple emotion attribute values.

[0127] In some embodiments, each of the multiple emotion attribute values indicates one emotion category among multiple emotion categories, and determining user evaluation indexes based on the multiple emotion attribute values includes: respectively determining, based on the multiple emotion attribute values, the proportion of the emotion attribute values indicating the multiple emotion categories among the multiple emotion attribute values; and using the weights of the respective multiple emotion categories to determine user evaluation indexes by weighting the proportions of the emotion attribute values of the multiple emotion categories.

[0128] In some embodiments, the sentiment attribute discrimination model includes a text extractor and an attention-based model, and determining the sentiment attribute values of a plurality of comments includes: for a given comment among the plurality of comments, using the text extractor to extract a text feature vector from the given comment; using the attention-based model to determine the attention weight of each of a plurality of text units in the given comment from the text feature vector; and determining the sentiment attribute value of the given comment based on the text feature vector and the determined attention weights.

[0129] In some embodiments, the value evaluation model includes at least a cross network and a deep network, and determining the value attribute value of a target brand includes: converting brand behavior data and user evaluation metrics into a combined feature vector; using the cross network to perform feature crossing on the combined feature vector representation to obtain a first feature; using the deep network to perform multiple feature extractions on the combined feature vector representation to obtain a second feature; and determining the value attribute value of the target brand based on the first feature and the second feature.

[0130] In some embodiments, the brand behavior data includes multiple types of brand behavior data; and converting the brand behavior data and user evaluation metrics into a combined feature vector includes: separately converting the multiple types of brand behavior data into multiple behavior feature vectors; classifying the user evaluation metrics and the multiple behavior feature vectors into a continuous feature set and a discrete feature set, with the user evaluation metrics being assigned to the continuous feature set; converting the feature vectors in the discrete feature set into feature vectors with a dense vector representation; and combining the continuous feature set and the feature vectors with the dense vector representation into the combined feature vector.

[0131] In some embodiments, the training of the value evaluation model includes: in a first training phase, training the value evaluation model using a first training data set, the first training data set including the brand behavior data and user evaluation data of a plurality of first sample brands and their labeled value attribute values; using the value evaluation model trained in the first training phase to determine a plurality of value evaluation results corresponding to a plurality of second sample brands based on the brand behavior data and user evaluation data of the plurality of second sample brands, the value attribute values of the plurality of second sample brands not being labeled; selecting at least one second sample brand to be labeled from the plurality of second sample brands based on the plurality of value evaluation results; obtaining the labeled value attribute values of the at least one second sample brand to obtain a second labeled training data set, the second labeled training data set including the brand behavior data and user evaluation data of the at least one second sample brand and their labeled value attribute values; and in a second training phase, using the second labeled training data set to update and train the value evaluation model.

[0132] In some embodiments, each of the multiple value evaluation results indicates multiple probabilities that the corresponding second sample brand belongs to multiple value categories, and selecting at least one second sample brand to be labeled from the multiple second sample brands includes: for each of the multiple second sample brands, if the difference degree between at least two of the multiple probabilities indicated by the value evaluation result corresponding to the second sample brand is lower than a preset threshold, selecting the second sample brand.

[0133] In some embodiments, the brand behavior data associated with the target brand includes at least one of the following: channel information associated with the target brand, volume information associated with the target brand, popularity information associated with the target brand, network metrics associated with the target brand, and user groups associated with the target brand.

[0134] In some embodiments, obtaining brand behavior data associated with the target brand and user evaluation data for the target brand includes: determining user selections in at least one of multiple industries related to the target brand; and obtaining brand behavior data and user evaluation data of the target brand within at least one industry; and the value attribute value of the target brand indicates the value attribute value of the target brand within the at least one industry.

[0135] The units included in apparatus 900 can be implemented in various ways, including software, hardware, firmware, or any combination thereof. In some embodiments, one or more units can be implemented using software and / or firmware, such as machine-executable instructions stored on a storage medium. In addition to or as an alternative to the machine-executable instructions, some or all of the units in apparatus 900 can be implemented at least in part by one or more hardware logic components. By way of example and not limitation, exemplary types of hardware logic components that can be used include field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0136] Figure 10 A block diagram of an electronic device 1000 in which one or more embodiments of the present disclosure can be implemented is shown. It should be understood that Figure 10 The illustrated electronic device 1000 is merely exemplary and should not impose any limitation on the functions and scope of the embodiments described herein. Figure 10 The illustrated electronic device 1000 can be used to implement Figure 1 the model application device 120 or the model training device 125.

[0137] As Figure 10As shown, the electronic device 1000 is in the form of a general-purpose electronic device. The components of the electronic device 1000 may include, but are not limited to, one or more processors or processing units 1010, a memory 1020, a storage device 1030, one or more communication units 1040, one or more input devices 1050, and one or more output devices 1060. The processing unit 1010 may be an actual or virtual processor and be capable of performing various processes according to the programs stored in the memory 1020. In a multi-processor system, multiple processing units execute computer-executable instructions in parallel to improve the parallel processing ability of the electronic device 1000.

[0138] The electronic device 1000 generally includes multiple computer storage media. Such media can be any available media accessible to the electronic device 1000, including but not limited to volatile and non-volatile media, removable and non-removable media. The memory 1020 may be a volatile memory (such as registers, caches, random access memory (RAM)), a non-volatile memory (such as read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory), or some combination thereof. The storage device 1030 may be a removable or non-removable medium and may include machine-readable media, such as a flash drive, a magnetic disk, or any other medium that can be used to store information and / or data (such as training data for training) and can be accessed within the electronic device 1000.

[0139] The electronic device 1000 may further include additional removable / non-removable, volatile / non-volatile storage media. Although not shown in Figure 10 a disk drive for reading from or writing to a removable, non-volatile magnetic disk (such as a "floppy disk") and an optical disk drive for reading from or writing to a removable, non-volatile optical disk may be provided. In these cases, each drive may be connected to a bus (not shown) by one or more data media interfaces. The memory 1020 may include a computer program product 1025 having one or more program modules configured to perform the various methods or actions of the various embodiments of the present disclosure.

[0140] The communication unit 1040 enables communication with other electronic devices through a communication medium. Additionally, the functions of the components of the electronic device 1000 may be implemented by a single computing cluster or multiple computer machines that are capable of communicating through a communication connection. Thus, the electronic device 1000 may operate in a networked environment using a logical connection with one or more other servers, network personal computers (PCs), or another network node.

[0141] The input device 1050 can be one or more input devices, such as a mouse, a keyboard, a trackball, etc. The output device 1060 can be one or more output devices, such as a display, a speaker, a printer, etc. The electronic device 1000 can also communicate with one or more external devices (not shown) as needed through the communication unit 1040. The external devices are such as a storage device, a display device, etc., communicate with one or more devices that enable a user to interact with the electronic device 1000, or communicate with any device (such as a network card, a modem, etc.) that enables the electronic device 1000 to communicate with one or more other electronic devices. Such communication can be performed via an input / output (I / O) interface (not shown).

[0142] According to an exemplary implementation of the present disclosure, there is provided a computer-readable storage medium having one or more computer instructions stored thereon, wherein the one or more computer instructions are executed by a processor to implement the method described above. According to an exemplary implementation of the present disclosure, there is also provided a computer program product. The computer program product is tangibly stored on a non-transitory computer-readable medium and includes computer-executable instructions, and the computer-executable instructions are executed by a processor to implement the method described above.

[0143] Aspects of the present disclosure are described herein with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the present disclosure. It should be understood that each block of the flowcharts and / or block diagrams, and the combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer-readable program instructions.

[0144] These computer-readable program instructions can be provided to a processing unit of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine such that when these instructions are executed by the processing unit of the computer or other programmable data processing device, a device is produced that implements the functions / actions specified in one or more blocks of the flowcharts and / or block diagrams. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause a computer, a programmable data processing device, and / or other devices to work in a specific manner. Thus, the computer-readable medium storing the instructions includes a manufactured article that includes instructions for implementing various aspects of the functions / actions specified in one or more blocks of the flowcharts and / or block diagrams.

[0145] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices, causing a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other devices to generate a computer-implemented process such that the instructions executed on the computer, other programmable data processing apparatus, or other devices implement the functions / acts specified in one or more boxes of the flowchart and / or block diagram.

[0146] The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various implementations of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a portion of an instruction, which contains one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions noted in the blocks may occur out of the order noted in the figures. For example, two consecutive blocks may in fact be executed substantially in parallel, or they may sometimes be executed in the reverse order, depending on the functionality involved. It should also be noted that each block of the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented by a dedicated hardware-based system that performs the specified functions or acts, or by a combination of dedicated hardware and computer instructions.

[0147] The various implementations of the present disclosure have been described above. The above description is exemplary, not exhaustive, and is not limited to the disclosed implementations. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described implementations. The choice of terms used herein is intended to best explain the principles of the implementations, the practical application, or improvements made to the technology in the marketplace, or to enable other ordinary skilled artisans in the art to understand the implementations disclosed herein.

Claims

1. A method for brand value evaluation, comprising: obtaining brand behavior data associated with a target brand and user evaluation data for the target brand; determining user evaluation indicators for the target brand based on the user evaluation data; and using a trained value evaluation model to determine a value attribute value of the target brand based on the brand behavior data and the user evaluation indicators of the target brand, the value evaluation model being trained to represent the association between the brand behavior data and user evaluation indicators of a brand and the value attribute value of the brand.

2. The method according to claim 1, wherein the user evaluation data includes multiple comments of multiple users on the target brand, and wherein determining the user evaluation indicators for the target brand based on the user evaluation data comprises: using an emotion attribute discrimination model to respectively determine multiple emotion attribute values of the multiple comments; and determining the user evaluation indicators based on the multiple emotion attribute values.

3. The method according to claim 2, wherein each of the multiple emotion attribute values indicates one emotion category among multiple emotion categories, and determining the user evaluation indicators based on the multiple emotion attribute values comprises: respectively determining, based on the multiple emotion attribute values, the proportions of the emotion attribute values indicating the multiple emotion categories among the multiple emotion attribute values; and using the weights of the multiple emotion categories respectively, and determining the user evaluation indicators by weighting the proportions of the emotion attribute values of the multiple emotion categories.

4. The method according to claim 2, wherein the emotion attribute discrimination model includes a text extractor and an attention-based model, and wherein determining the multiple emotion attribute values of the multiple comments comprises: for a given comment among the multiple comments, using the text extractor to extract a text feature vector from the given comment; using the attention-based model to determine the attention weights of multiple text units in the given comment from the text feature vector; and determining the emotion attribute value of the given comment based on the text feature vector and the determined attention weights.

5. The method according to claim 1, wherein the value evaluation model includes at least a cross network and a deep network, and wherein determining the value attribute value of the target brand comprises: converting the brand behavior data and the user evaluation indicators into a combined feature vector; using the cross network to perform feature crossing on the combined feature vector representation to obtain a first feature; using the deep network to perform multiple feature extractions on the combined feature vector representation to obtain a second feature; and determining the value attribute value of the target brand based on the first feature and the second feature.

6. The method according to claim 5, wherein the brand behavior data includes multiple types of brand behavior data; and wherein converting the brand behavior data and the user evaluation indicators into a combined feature vector comprises: respectively converting the multiple types of brand behavior data into multiple behavior feature vectors; Classify the user evaluation metrics and the multiple behavioral feature vectors into a continuous feature set and a discrete feature set, with the user evaluation metrics being partitioned into the continuous feature set; Convert the feature vectors in the discrete feature set into feature vectors represented by dense vectors; and Merge the continuous feature set and the feature vectors represented by dense vectors into the combined feature vector.

7. The method according to any one of claims 1 - 6, wherein the training of the value evaluation model comprises: In a first training phase, train the value evaluation model using a first training data set, the first training data set including brand behavior data, user evaluation data, and marked value attribute values of multiple first sample brands respectively; Using the value evaluation model trained in the first training phase, determine multiple value evaluation results corresponding to the multiple second sample brands based on the brand behavior data and user evaluation data of the multiple second sample brands, the value attribute values of the multiple second sample brands not being marked; Based on the multiple value evaluation results, select at least one second sample brand to be marked from the multiple second sample brands; Obtain the marked value attribute values of the at least one second sample brand respectively, to obtain a second marked training data set, the second marked training data set including the brand behavior data, user evaluation data, and marked value attribute values of the at least one second sample brand respectively; and In a second training phase, update and train the value evaluation model using the second marked training data set.

8. The method according to claim 7, wherein each value evaluation result among the multiple value evaluation results indicates multiple probabilities that the corresponding second sample brand belongs to multiple value categories, and wherein selecting at least one second sample brand to be marked from the multiple second sample brands comprises: For each second sample brand among the multiple second sample brands, if the difference degree between at least two of the multiple probabilities indicated by the value evaluation result corresponding to this second sample brand is lower than a preset threshold, select this second sample brand.

9. The method according to any one of claims 1 - 8, wherein the brand behavior data associated with the target brand includes at least one of the following: Channel information associated with the target brand, Volume information associated with the target brand, Popularity information associated with the target brand, Network metrics associated with the target brand, and User groups associated with the target brand.

10. The method according to claim 1, wherein obtaining brand behavior data associated with a target brand and user evaluation data for the target brand comprises: Determine the user selection in at least one industry related to the target brand; and Obtain the brand behavior data and user evaluation data of the target brand within the at least one industry; and wherein the value attribute value of the target brand indicates the value attribute value of the target brand within the at least one industry.

11. An apparatus for brand value evaluation, comprising: An acquisition module, configured to acquire brand behavior data associated with a target brand and user evaluation data for the target brand; A user evaluation index determination module, configured to determine a user evaluation index for the target brand based on the user evaluation data; And A value attribute value determination module, configured to use a trained value evaluation model to determine a value attribute value of the target brand based on the brand behavior data and the user evaluation index of the target brand, where the value evaluation model is trained to represent the association relationship between the brand behavior data and user evaluation index of a brand and the value attribute value of the brand.

12. An electronic device, Comprising: At least one processing unit; And At least one memory, the at least one memory being coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit, the instructions, when executed by the at least one processing unit, causing the electronic device to execute the method according to any one of claims 1 to 10.

13. A computer-readable storage medium, having stored thereon a computer program, the computer program being executable by a processor to implement the method according to any one of claims 1 to 10.

Images