A method, apparatus and storage medium for predicting commodity rating distribution

Abstract

This invention discloses a method, apparatus, and storage medium for predicting product rating distributions. The method includes: acquiring a dataset; constructing a local neighborhood graph for each data point based on its distance to neighboring points; using spectral embedding, utilizing the eigenvalues ​​and eigenvectors of the graph's Laplacian matrix to find the optimal low-dimensional embedding, thus preserving the local distance relationships between data points in the high-dimensional space; progressively adjusting the positions of data points in the low-dimensional space using stochastic gradient descent to minimize the difference between the graphs in the input space and the low-dimensional space, obtaining a low-dimensional representation of the original input space; and using the low-dimensional representation of the original input space and the product label distribution as input to a label distribution support vector regressor to achieve accurate prediction of the product rating distribution. This invention can effectively handle high-dimensional, multi-label datasets, significantly reducing computational costs and improving prediction accuracy and efficiency, and can be widely applied in the field of computer data mining.

Description

Technical Field

[0001] This invention relates to the field of computer data mining, and in particular to a method, apparatus and storage medium for predicting the distribution of product ratings. Background Technology

[0002] With accelerating globalization, the retail industry has become a multi-billion dollar international business sector. This rapid growth is accompanied by thousands of new products entering the market each year, but their success rates vary considerably. From manufacturers' perspective, rising production costs and intense market competition exacerbate investment risks. Consumers, on the other hand, face a different challenge: the proliferation of advertising and marketing campaigns makes it difficult for them to choose products that meet their individual needs. Against this backdrop, accurately predicting consumer attitudes and preferences for upcoming retail products, even before the product design stage, has become a pressing issue for both parties.

[0003] Advances in machine learning technology have offered an effective solution to this challenge. By utilizing machine learning-based predictive models, we can now make early predictions about consumer reactions to products, thus providing a scientific basis for product adjustments and market strategy development. In this regard, consumer ratings of products are particularly important, serving as a key indicator reflecting a product's potential market acceptance and purchasing potential.

[0004] However, traditional forecasting methods often focus on evaluating consumers' average rating scores for products using mathematical statistical methods, using this as a standard to measure product quality. While this method can reflect consumers' overall perceived quality of products, it ignores the distributional nature of product ratings, thus missing rich semantic information. Summary of the Invention

[0005] In order to at least partially solve one of the technical problems existing in the prior art, the present invention aims to provide a method, apparatus and storage medium for predicting product rating distribution based on dimensionality reduction and label distribution learning.

[0006] The first technical solution adopted in this invention is:

[0007] A method for predicting the distribution of product ratings includes the following steps:

[0008] Data acquisition and preprocessing: The acquired packaging images and product attributes of a certain type of product are preprocessed to obtain a preprocessed dataset;

[0009] Constructing a fuzzy weighted undirected graph: Based on the preprocessed dataset, for each data point, a local neighborhood graph is constructed according to its distance from neighboring points; this local neighborhood graph is a fuzzy weighted undirected graph, where nodes represent data points and the weights of edges reflect the similarity or proximity between points;

[0010] Constructing an initial low-dimensional representation: Through spectral embedding, the optimal low-dimensional embedding is found by utilizing the eigenvalues ​​and eigenvectors of the Laplacian matrix of the graph to preserve the local distance relationships between data points in the high-dimensional space;

[0011] Optimize weighted fuzzy cross-entropy: Use weighted fuzzy cross-entropy as the loss function and use stochastic gradient descent to gradually adjust the positions of data points in the low-dimensional space to minimize the difference between the graphs in the input space and the low-dimensional space, and obtain a low-dimensional representation of the original input space.

[0012] Predicting the product rating distribution: Using the low-dimensional representation of the original input space and the product label distribution as input to the label distribution support vector regressor, accurate prediction of the product rating distribution can be achieved.

[0013] Furthermore, the preprocessing of the obtained packaging images and product attributes of a certain type of product to obtain a preprocessed dataset includes:

[0014] Color histograms are extracted to form color feature vectors, product attributes are acquired and encoded, and all features are normalized to obtain the final dataset X = {x1, x2, ..., x...}. n}

[0015] Furthermore, the product rating distribution prediction method also includes the following steps:

[0016] The distribution of user ratings for products is obtained by calculating the frequency distribution of ratings:

[0017] Define a set of rating levels S = {s1, s2, ..., s} m}, where m is the total number of rating levels;

[0018] For each item x i There is a rating set Where n j It is product p j The total number of ratings;

[0019] Rating Rank s i In product p j The frequency of the rating distribution f ji The calculation formula is:

[0020]

[0021] Among them, F ji Indicates rating level s i x% of goods i The proportion of the overall rating, and thus the product x i The rating distribution is represented by a vector f. j =(f j1 f j2 , ..., f jm This intuitively reflects the distribution of each rating level F.

[0022] Furthermore, based on the preprocessed dataset, for each data point, a local neighborhood graph is constructed according to its distance to neighboring points, including:

[0023] In the dataset X = {x1, x2, ..., x...} n In}, each point x i Find its k nearest neighbors T using K-nearest neighbor search. i ={t i1 , ..., t il , ..., t ik}, and calculate x i With each neighbor t il The distance d between il This forms a set D. i ={d i1 , ..., d il , ..., d ik}, containing x i The distance between each of its k nearest neighbors;

[0024] After calculating the distance between each point in the dataset and its k nearest neighbors, a fuzzy simplex is constructed to represent the similarity between objects, and a local neighborhood graph is constructed based on the similarity.

[0025] Furthermore, the construction of a fuzzy simplex set to represent the similarity between objects, and the construction of a local neighborhood graph based on the similarity, includes:

[0026] For each object, determine a minimum distance ρ. i , ρ i D is the set of distances D from the object to its nearest neighbors. i The minimum value in;

[0027] Find a σ using binary search. i This makes the following formula true:

[0028]

[0029] Using a defined ρ i and σi A vector was constructed for each object. Its elements μ ij The similarity between objects is calculated using the following formula:

[0030]

[0031] If two objects are not neighbors, the similarity is set to 0; thus, for each object x... i Corresponding to a sparse vector It encodes the fuzzy similarity between itself and other objects in the dataset, constructing a sparse weighted adjacency matrix. This represents a fuzzy weighted directed graph that shows pairwise similarity between encoded objects.

[0032] The asymmetric matrix M is symmetricized using the following formula to obtain a symmetric matrix representing the pairwise similarity between objects:

[0033] μ ij ←μ ij +μ ji -μ ij μ ji

[0034] Furthermore, the step of finding the optimal low-dimensional embedding through spectral embedding, utilizing the eigenvalues ​​and eigenvectors of the graph's Laplacian matrix, includes:

[0035] Transform the adjacency matrix M into a transformation matrix C, where each element c ij Let represent the probability of transition from point i to point j; this transition involves a standardization step such that the sum of the elements in each row is 1, i.e., ∑ j c ij =1;

[0036] Perform eigenvalue decomposition on the transformation matrix C to find its m largest non-zero eigenvalues ​​and their corresponding eigenvectors {λ1, λ2, ..., λ3}. m}and These eigenvectors constitute the initial embedding of the data in m-dimensional space, forming the initial embedding matrix.

[0037] Furthermore, the method of using weighted fuzzy cross-entropy as the loss function and using stochastic gradient descent to progressively adjust the positions of data points in the low-dimensional space to minimize the difference between the graphs in the input space and the low-dimensional space, thereby obtaining a low-dimensional representation of the original input space, includes:

[0038] Based on the original high-dimensional dataset X, an initial embedding matrix is ​​calculated. Let represent the initial positions of n objects in m-dimensional space; weighted fuzzy cross-entropy is used as the loss function, and stochastic gradient descent is employed to optimize it; the loss function is defined as:

[0039]

[0040] Where M is a symmetric adjacency matrix, encoding pairwise fuzzy similarity between objects; μ ij μ ik It is the fuzzy similarity between objects in the original dataset; v ij It is the fuzzy similarity between objects in a low-dimensional space;

[0041] The similarity v in the low-dimensional space is calculated using the following formula. ij :

[0042]

[0043] Where, d ij It is the Euclidean distance between two objects in low-dimensional space, where a and b are coefficients chosen through fitting; by iteratively updating the Y matrix, the new coordinates of the data points in low-dimensional space are obtained, and the data is divided into training and test sets.

[0044] Furthermore, the label distribution support vector regressor is trained in the following manner:

[0045] Learn the mapping function from features to score distribution using the following objective function:

[0046]

[0047] in, Represents the weight matrix. This represents the bias vector, where W and b are model parameters. j This represents the transpose of the j-th row of w, while L(r) i ) is the loss function defined for the i-th instance, and is defined as follows:

[0048]

[0049]

[0050]

[0051] The training and test sets are used as inputs to a label distribution support vector regressor to predict the score distribution of goods.

[0052] Chebyshev distance, Clark distance, KL divergence, and cosine similarity were used as evaluation metrics.

[0053] The second technical solution adopted in this invention is:

[0054] A product rating distribution prediction device, comprising:

[0055] At least one processor;

[0056] At least one memory for storing at least one program;

[0057] When the at least one program is executed by the at least one processor, the at least one processor implements the method described above.

[0058] The third technical solution adopted in this invention is:

[0059] A computer-readable storage medium storing a processor-executable program, which, when executed by a processor, performs the method described above.

[0060] The beneficial effects of this invention are as follows: This invention preprocesses commodity data and constructs a fuzzy weighted undirected graph, uses spectral embedding to obtain a low-dimensional representation of the data, optimizes the low-dimensional representation using weighted fuzzy cross-entropy, and finally uses a label distribution support vector regressor to predict the commodity rating distribution. This invention can effectively handle high-dimensional, multi-label datasets, which not only significantly reduces computational costs but also improves the accuracy and efficiency of prediction. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following description is provided with accompanying drawings of the relevant technical solutions in the embodiments of the present invention or the prior art. It should be understood that the accompanying drawings described below are only for the purpose of clearly illustrating some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 This is a flowchart illustrating the steps of a product rating distribution prediction method in an embodiment of the present invention. Detailed Implementation

[0063] The embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. The step numbers in the following embodiments are set only for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0064] In the description of this invention, it should be understood that the orientation descriptions, such as up, down, front, back, left, right, etc., are based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0065] In the description of this invention, "several" means one or more, "multiple" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. If "first" or "second" is used, it is only for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features. Furthermore, "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0066] In the description of this invention, unless otherwise explicitly defined, terms such as "set up," "install," and "connect" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0067] Related consumer psychology research shows that analyzing the distribution of consumer ratings for a particular product can provide deeper insights. For example, a sharp distribution of ratings suggests consistency in product ratings, while a broad distribution indicates wide-ranging disagreements in opinions about the product. This distributed analysis provides manufacturers with more precise and multi-dimensional market feedback, enabling them to develop more targeted marketing strategies based on the specific characteristics of consumer feedback, thereby enhancing product market awareness and sales potential.

[0068] like Figure 1 As shown, this embodiment provides a novel product rating distribution prediction method based on label distribution learning technology. It preprocesses product data and constructs a fuzzy weighted undirected graph, then uses spectral embedding to obtain a low-dimensional representation of the data. Further, weighted fuzzy cross-entropy is used to optimize the low-dimensional representation, and finally, a label distribution support vector regressor is used to predict the product rating distribution. This method can effectively handle high-dimensional, multi-label datasets, significantly reducing computational costs and improving prediction accuracy and efficiency. This advancement is significant for optimizing inventory management, improving customer satisfaction, and reducing waste, providing a more accurate and efficient solution for product rating distribution prediction. The method specifically includes the following steps:

[0069] S1. Data Acquisition and Preprocessing: Preprocess the packaging images, product attribute data, and product ratings of a certain category of goods to obtain the processed dataset.

[0070] In some embodiments, step S1 specifically includes the following steps:

[0071] S11. Before model training, the real-world product dataset needs to be processed. First, color histograms are extracted from packaging images of specific product categories, constructing color feature vectors using 256 color channels. This step considers the potential impact of packaging color on consumer emotions, aiming to form a high-dimensional feature representation by quantifying the color distribution of the images. Second, product attribute features are obtained, specifically including: product category, brand, price, packaging size, and packaging geometry; one-hot encoding is used to encode the product category, brand, packaging size, and packaging geometry respectively. Finally, a normalization strategy is adopted for all features to standardize the data volume, improve the convergence speed of the algorithm, and enhance the model's generalization ability. This method, by adjusting the data distribution, ensures the fairness of comparisons and weight allocation between different features, helping to optimize model training efficiency and prediction accuracy.

[0072] S12. Calculate the rating distribution for each product, and define the set of rating levels as S = {s1, s2, ..., s}. m}, where m is the total number of rating levels. For products Its rating set is denoted as n j For product p j The total number of ratings.

[0073] Rating Rank s i In product p j The frequency of the rating distribution f ji The calculation formula is:

[0074]

[0075] The formula calculates f. ji Indicates rating level s i Commodities The proportion of the overall rating. Therefore, the commodity... The rating distribution can be represented as a vector. It intuitively reflects the distribution of each rating level.

[0076] S2. Constructing a fuzzy weighted undirected graph captures local neighborhood information between data points and also considers their similarity in the form of a fuzzy metric.

[0077] In some embodiments, step S2 specifically includes the following steps:

[0078] S21. For the dataset Each point in We perform a K-nearest neighbor search to find its k nearest neighbors. Assumption For T i For each neighbor found in the set, calculate and scalar distance d between il scalar d il The value is calculated as follows:

[0079]

[0080] Where i is the index of an object in dataset X; l is the index of one of the k nearest neighbors of the i-th object; h represents the vector. That is, the dimension of the i-th object; It is the i-th object and its value in T i The scalar distance between the l-th nearest neighbors in the set.

[0081] Therefore, for each object in dataset X We have defined a set D. i ={d i1 , ..., d il ,,..,d ik}, this set contains The distance between each of its k nearest neighbors.

[0082] S22, after calculation After calculating the distances to each of its k nearest neighbors, a fuzzy simplex set is constructed, represented as a vector. Where n represents the number of objects in the original high-dimensional dataset. To construct each i-th object... Vector, algorithm to find ρ i ∈D i , making Next, perform a binary search to find σ that satisfies the following conditions. i :

[0083]

[0084] Where i is the ID of an object in dataset X; l is the ID of the nearest neighbor of the i-th object; k represents the number of nearest neighbors; It is the target variable; ρ i ∈D i It is an object Rather than in T containing k neighbors i The distance between the nearest neighbors in a set.

[0085] After determining ρ i And for each i-th object in the original high-dimensional dataset X Find σ that meets the conditions i Then, a sparse vector was constructed. vector Each j-th scalar component is represented by a fuzzy value indicating the similarity between the i-th and j-th objects in the X set. Assume i = {1, 2, ..., n} and j = {1, 2, ..., n}, where n represents the number of objects in the X set. If two objects... and If they're not neighbors, then... The j-th component μ of the vector ij It is set to 0. If two objects... and If they are neighbors, then μ ij The calculation method is as follows:

[0086]

[0087] Where i is the value being built The object number of the vector; j is the number of a possible neighbor of the i-th object in set X, ρ i It comes from D i Minimum distance between sets; d ij yes and The distance between them; The vector has dimensions n and μ. ij ∈[0,1].

[0088] The result is that for every object in X They all obtained a sparse vector It encodes the model-orientation similarity between the i-th object and every j-th object in the original high-dimensional dataset X. Given i = {1, 2, ..., n}, we construct a sparse weighted adjacency matrix. The n rows consist of n sparse mode lake vectors The weighted adjacency matrix M represents a fuzzy weighted directed graph that encodes pairwise similarities between objects in X. M is not symmetric.

[0089] The next step is to symmetricize the asymmetric matrix M according to the following formula:

[0090] μ ij ←μ ij +μ ji-μ ij μ ji

[0091] In matrix M, i and j represent row and column numbers, respectively. Note that μ ii and μ jj All are equal to 0. Therefore, the adjacency matrix M becomes a symmetric matrix. The construction of the fuzzy simplex is achieved by defining the fuzzy similarity between each data point and its neighbors. Specifically, by defining the fuzzy similarity between each data point x and its neighbors... i Assign a fuzzy metric μ to the relationship between its k nearest neighbors. ij In essence, we have defined a fuzzy relationship between points in a high-dimensional dataset. This fuzzy relationship is expressed through vectors. It is represented that each element μ ij It reflects point x i Other points x j The similarity between them. In this way, we not only capture the local neighborhood information between data points, but also consider their similarity in the form of a fuzzy metric, thus constructing a fuzzy simplex set.

[0092] This fuzzy simple collective is now in a sparse weighted adjacency matrix M, where the element μ ij and μ ji They represent the points from point x i To x j And from point x j To x i The fuzzy similarity is obtained by symmetrizing the asymmetric matrix M. This not only reflects a more balanced similarity relationship but also makes the structure of the fuzzy simplex more explicit, as it emphasizes the importance of the reciprocity between data points and ensures the bidirectional consistency of similarity. Through this method, we establish a fuzzy representation of complex data structures, effectively capturing the inherent structure and similarity patterns of the data.

[0093] S3. Construct the initial low-dimensional representation. First, transform the adjacency matrix M into a transformation matrix C, where each element c ij Let represent the probability of transitioning from point i to point j. This transition involves a standardization step, such that the sum of the elements in each row is 1, i.e., ∑ j c ij =1. Next, perform eigenvalue decomposition on the transformation matrix C to find its m largest non-zero eigenvalues ​​and corresponding eigenvectors {λ1, λ2, ..., λ3}. m}and These eigenvectors constitute the initial embedding of the data in m-dimensional space, forming the initial embedding matrix.

[0094] S4. Optimize the weighted fuzzy cross-entropy to obtain the initial embedding matrix. That is, after calculating the object from X in After determining the initial position in space, we begin using weighted fuzzy cross-entropy as the loss function and optimize it using stochastic gradient descent. The loss function is:

[0095]

[0096] in Represents a symmetric adjacency matrix containing fuzzy values, encoding pairwise similarities between high-dimensional objects from the set X; Representing low-dimensional space The representation of n objects; μ ij ∈[0,1] represents the fuzzy similarity between the i-th and j-th high-dimensional objects in the original X set; v ij ∈[0,1] represents a low-dimensional space The fuzzy similarity between the i-th and j-th objects. To determine the fuzzy similarity between the i-th and j-th objects. The low-dimensional space represented by the i-th and j-th rows of the matrix The pairwise similarity v between the i-th and j-th objects ij Use the following formula:

[0097]

[0098] Where d ij This represents the i-th and j-th objects represented by the rows in the Y matrix. and The scalar distance between them is calculated using the Euclidean distance formula. a and b are coefficients chosen using a nonlinear least squares fitting method, and d... min It is assumed that the hyperparameters are set within the range (0, 1):

[0099]

[0100] Finally, the iteratively updated Y matrix is ​​obtained, which represents the coordinates of points in the low-dimensional space. The data is then divided into training and test sets.

[0101] S5. Construct a label distribution support vector regressor. The score distribution F we obtain is modeled using element-wise sigmoid vectors as follows:

[0102]

[0103] Transform x in a nonlinear manner to a higher-dimensional feature space. In; among them, and These are the model parameters. We learn the mapping function from features to score distribution using the following objective function:

[0104]

[0105] in, Represents the weight matrix. This represents the bias vector, where W and b are model parameters. j This represents the transpose of the j-th row of W, while L(a) represents the transpose of the j-th row of W. i ′) is the loss function defined for the i-th instance, and is defined as:

[0106]

[0107]

[0108]

[0109] The training and test sets are used as inputs to a label distribution support vector regressor to predict the score distribution of items. Chebyshev distance, Clark distance, KL divergence, and cosine similarity are used as evaluation metrics, calculated as follows:

[0110] Chebyshev distance↓max i |p i -q i |

[0111] Clark distance↓

[0112] KL divergence↓

[0113] Cosine similarity↑

[0114] In summary, compared with the prior art, the present invention has at least the following advantages and beneficial effects:

[0115] (1) Compared to traditional marketing forecasting methods that primarily focus on average product rating scores, this invention provides a method for in-depth analysis of product rating distribution through machine learning technology. This method not only captures information about consumers' average perception of products, but more importantly, it reveals the distributional nature of product ratings, such as consistency or divergence, thus providing richer and more multi-dimensional market feedback. This enables manufacturers to develop more precise and targeted marketing strategies based on the specific characteristics of consumer feedback, significantly improving product market awareness and sales potential.

[0116] (2) This invention effectively improves the accuracy of product rating distribution prediction by combining dimensionality reduction technology and label distribution learning technology. Through preprocessing and comprehensive analysis of product packaging images, attribute data and product ratings, this method can more accurately predict the distribution of user ratings for products before they are put on the shelves, grasp consumer expectations and preferences, and gain insights into future market trends, providing enterprise decision-makers with more forward-looking product rating distribution prediction information.

[0117] (3) This invention employs fuzzy weighted undirected graph and spectral embedding techniques to optimize the processing efficiency of high-dimensional data and reduce computational costs. This method not only preserves the local and global distance relationships between data points and improves the generalization ability of the prediction model, but also effectively shortens the model training time through weighted fuzzy cross-entropy and stochastic gradient descent.

[0118] The present invention also provides a product rating distribution prediction device, comprising:

[0119] At least one processor;

[0120] At least one memory for storing at least one program;

[0121] When the at least one program is executed by the at least one processor, the at least one processor implements Figure 1 The method shown.

[0122] This embodiment of the product rating distribution prediction device can execute a product rating distribution prediction method provided in the method embodiment of the present invention, and can execute any combination of implementation steps of the method embodiment, and has the corresponding functions and beneficial effects of the method.

[0123] This application also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform... Figure 1 The method shown.

[0124] This embodiment also provides a storage medium storing instructions or programs that can execute the commodity rating distribution prediction method provided in the method embodiment of the present invention. When the instructions or programs are run, any combination of implementation steps of the method embodiment can be executed, and the method has the corresponding functions and beneficial effects.

[0125] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is altered and sub-operations described as part of a larger operation are executed independently.

[0126] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the described functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0127] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0128] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0129] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0130] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0131] In the foregoing description of this specification, references to terms such as "one embodiment," "another embodiment," or "some embodiments" indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0132] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0133] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for predicting the distribution of commodity ratings, characterized in that, Includes the following steps: The obtained packaging images and product attributes of a certain type of product are preprocessed to obtain a preprocessed dataset; Based on the preprocessed dataset, for each data point, a local neighborhood graph is constructed according to its distance to neighboring points; By using spectral embedding, the optimal low-dimensional embedding is found by utilizing the eigenvalues ​​and eigenvectors of the Laplacian matrix of the graph, in order to preserve the local distance relationships between data points in the high-dimensional space. We use weighted fuzzy cross-entropy as the loss function and stochastic gradient descent to gradually adjust the positions of data points in the low-dimensional space in order to minimize the difference between the graphs in the input space and the low-dimensional space and obtain a low-dimensional representation of the original input space. By using the low-dimensional representation of the original input space and the product label distribution as input to the label distribution support vector regressor, accurate prediction of the product rating distribution can be achieved.

2. The commodity rating distribution prediction method according to claim 1, characterized in that, The process of preprocessing the obtained packaging images and product attributes of a certain type of product to obtain a preprocessed dataset includes: Color histograms are extracted to form color feature vectors, product attributes are acquired and encoded, and all features are normalized to obtain the final dataset. .

3. The product rating distribution prediction method according to claim 1, characterized in that, The commodity rating distribution prediction method also includes the following steps: The distribution of user ratings for products is obtained by calculating the frequency distribution of ratings: Define a set of rating levels ,in It represents the total number of rating levels; For each product There is a rating set ,in It is a commodity The total number of ratings; Rating In the goods Rating distribution frequency The calculation formula is: in, Indicates rating level Commodities The proportion of the overall rating, and thus the product The rating distribution is represented as a vector. This intuitively reflects the distribution of each rating level. The situation.

4. The commodity rating distribution prediction method according to claim 1, characterized in that, Based on the preprocessed dataset, for each data point, a local neighborhood graph is constructed according to its distance to neighboring points, including: In the dataset In, each point pass Nearest neighbor search found it The nearest neighbor and calculate With each neighbor Distance between , forming a set ,Include With The distance between the nearest neighbors; Each point in the dataset is compared with its... After calculating the distances of the nearest neighbors, a fuzzy simplex is constructed to represent the similarity between objects, and a local neighborhood graph is built based on the similarity.

5. The product rating distribution prediction method according to claim 4, characterized in that, The construction of a fuzzy simplex set to represent the similarity between objects, and the construction of a local neighborhood graph based on the similarity, includes: For each object, determine a minimum distance. , It is the set of distances from the object to its nearest neighbor. The minimum value in; A binary search was used to find one. This makes the following formula true: Use a defined and A vector was constructed for each object. Its elements The similarity between objects is calculated using the following formula: If two objects are not neighbors, the similarity is set to 0; thus, each object... Corresponding to a sparse vector It encodes the fuzzy similarity between itself and other objects in the dataset, constructing a sparse weighted adjacency matrix. This represents a fuzzy weighted directed graph that encodes pairwise similarities between objects; The following formula applies to asymmetric matrices. After symmetrization, we obtain a symmetric matrix representing the pairwise similarity between objects: 。 6. The product rating distribution prediction method according to claim 1, characterized in that, The method of finding the optimal low-dimensional embedding through spectral embedding, utilizing the eigenvalues ​​and eigenvectors of the Laplacian matrix of the graph, includes: Adjacency matrix Transform into a transformation matrix C, where each element express Time The transition probability; Perform eigenvalue decomposition on the transformation matrix C to find its The largest non-zero eigenvalues ​​and their corresponding eigenvectors and These feature vectors constitute the data in The initial embeddings in the dimensional space form the initial embedding matrix. .

7. The product rating distribution prediction method according to claim 1, characterized in that, The method of using weighted fuzzy cross-entropy as the loss function and employing stochastic gradient descent to progressively adjust the positions of data points in the low-dimensional space to minimize the difference between the graphs in the input space and the low-dimensional space, thereby obtaining a low-dimensional representation of the original input space, includes: Based on the original high-dimensional dataset An initial embedding matrix is ​​calculated. ,express An object in The initial position in dimensional space; weighted fuzzy cross-entropy is used as the loss function, and it is optimized using stochastic gradient descent; the loss function is defined as: in, It is a symmetric adjacency matrix that encodes pairwise fuzzy similarity between objects; , It is the fuzzy similarity between objects in the original dataset; It is the fuzzy similarity between objects in a low-dimensional space; Similarity in low-dimensional space is calculated using the following formula. : in, It is the Euclidean distance between two objects in a low-dimensional space. and These are coefficients; updated iteratively. The matrix is ​​used to obtain the new coordinates of the data points in the low-dimensional space, and the data is divided into training and test sets.

8. The product rating distribution prediction method according to claim 1, characterized in that, The label distribution support vector regressor is trained in the following way: Learn the mapping function from features to score distribution using the following objective function: in, Represents the weight matrix. This represents the bias vector. It means The The transpose of a row, and It is aimed at the first The loss function defined for each instance; The training and test sets are used as inputs to a label distribution support vector regressor to predict the score distribution of items.

9. A commodity rating distribution prediction device, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method of any one of claims 1-8.

10. A computer-readable storage medium storing a processor-executable program, characterized in that, The processor-executable program, when executed by the processor, is used to perform the method as described in any one of claims 1-8.

Images