Multi-view clustering integrated sales factor multi-preference decision evaluation method and system based on data reconstruction
By employing a multi-view clustering ensemble method, the problem of difficulty in mining the diversity and complexity of sales factor data in existing technologies is solved, enabling accurate positioning of key sales factors and reliable sales decision support.
Patent Information
- Application Number
- CN202511078031.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-01
- Publication Date
- 2025-11-11
AI Technical Summary
Existing technologies are insufficient to fully consider consumers' actual needs and cannot fully explore the diversity and complexity of sales factor data, making it difficult to accurately identify key sales factors.
A multi-view clustering integration method based on data reconstruction is adopted. This method generates multiple view groups by standardizing and preprocessing multi-preference decision data of sales factors, granularizing data point selection, and permuting and combining them. A sparse sub-bipartite graph is constructed by a multi-view anchor point selection strategy of view group collaboration. Finally, spectral clustering is performed to obtain the final clustering result.
It improves the accuracy and robustness of data analysis, breaks through the information limitations of a single view, comprehensively covers the correlation between sales factors in multiple scenarios, achieves accurate positioning of key sales factors, and provides a reliable basis for sales decision-making.
Smart Images
Figure CN120931312A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of sales factor evaluation technology, specifically relating to a sales factor multi-preference decision evaluation method and system based on data reconstruction and multi-view clustering integration. Background Technology
[0002] With continuous technological advancements and diversified consumer demands, sales factors are playing an increasingly important role in the business world. These factors profoundly impact people's daily lives, consumption experiences, and socio-economic development. The supply of goods, price levels, and consumption habits directly influence people's lifestyles and quality of life. Consumers' demands for product quality, price, and shopping experience are constantly rising, prompting businesses to deeply understand consumer psychology and behavior and develop more targeted sales strategies. However, numerous factors influence sales, and identifying the factors that consumers care about most is crucial for ensuring stable and sustainable sales growth. For example, understanding which factors have the greatest impact on sales allows businesses to more accurately allocate resources (such as marketing budgets, manpower, and time) to these key areas. This targeted resource allocation maximizes resource utilization efficiency and avoids waste and dispersion. Analyzing the most important factors simplifies the decision-making process, making decisions faster and more efficient. The analysis results of the most important sales factors not only guide current sales activities but also provide important basis for future strategic planning.
[0003] Sales factor selection plays a crucial role in business decision-making, helping companies better understand and utilize various sales factors to optimize sales strategies, improve performance, and achieve business goals. Evaluating and selecting multiple sales factors enables more refined decision-making, identifying which factors are most critical to sales performance, thus improving the relevance and accuracy of decisions. It helps companies effectively allocate resources, avoid waste, and concentrate resources on the most impactful areas. It helps companies consider sales factors comprehensively from multiple perspectives, making decisions more objective and scientific. Data-driven decision-making helps reduce subjective bias and improve the accuracy of judgments. By selecting key sales factors, companies can optimize sales strategies, better meet customer needs, and enhance market competitiveness. A deeper understanding of customer preferences and behaviors helps in customizing personalized sales solutions. It reduces decision-making errors and improves operational efficiency. In summary, sales factor selection not only helps companies better understand the market environment and customer needs but also optimizes resource allocation, improves decision-making effectiveness, and drives business growth, which is crucial for business development.
[0004] Traditional sales factor analysis methods, primarily based on historical data or simple surveys, struggle to fully consider real-world consumer needs such as preferences. While companies employ various methods to select sales factors, this area still faces challenges that prevent them from achieving their intended purpose. Effective sales factor selection requires extracting more reliable and useful factors from multi-preference decision-making data, identifying the most crucial factors to optimize sales strategies and make more refined decisions. Furthermore, the sheer volume of data in sales factor analysis can lead to noise or missing information in single-view data. Current technologies often employ clustering methods to reduce data errors; however, most existing clustering methods rely on a single clustering approach for single-view data, which has limitations. A single clustering method based solely on a single view of data may not fully capture the diversity and complexity of the data. Summary of the Invention
[0005] This invention provides a sales factor multi-preference decision evaluation method and system based on data reconstruction and multi-view clustering integration, in order to solve the technical problems in the prior art that make it difficult to fully consider the actual needs of consumers, to fully explore the diversity and complexity of data, and thus difficult to accurately discover key sales factors.
[0006] To achieve the above objectives, the present invention adopts the following technical solution: A multi-view clustering ensemble method for evaluating sales factors based on data reconstruction and multi-preference decision-making includes the following steps: Acquire multi-view data on sales factors, preferences, and decisions; arrange and combine each view data to generate multiple view groups. Data reconstruction is performed on the view units in each view group to obtain multiple sparse sub-bipartite graphs. These multiple sparse sub-bipartite graphs are then merged to obtain a joint sparse bipartite graph of the views. The view joint sparse bipartite graph is partitioned and integrated to obtain the final clustering result; based on the final clustering result, sales factors are evaluated.
[0007] The acquisition of multi-preference decision data based on sales factors involves permuting and combining data from each view to generate multiple view groups. Specifically, the acquired multi-preference decision data is preprocessed using standardization. Then, for each view in the processed multi-preference decision data, representative sample points are selected sequentially using a granular data selection method. This process can be divided into two main stages: the first stage reduces the data volume while preserving the data's distribution characteristics through sample point selection; the second stage, based on the first stage, reduces the data dimensionality by dimensionality reduction while preserving the original key features of the data. That is, the multi-preference decision data based on sales factors is transformed from... Become , in This represents the first factor in the multi-preference decision dataset related to sales factors. One view, This represents the first of the multiple preference decision datasets for sales factors after point selection. One view, The total number of views for the multi-preference decision data of sales factors is determined; then, each view of the multi-preference decision data of sales factors after selection is used to generate multiple view groups through permutation and combination.
[0008] The acquired sales factor preference decision data is subjected to standardized preprocessing as follows:
[0009] in, For initial sales factors, multiple preference decision data, To standardize the preprocessed sales factor preference decision data, All of these are known constants.
[0010] The sales factor multi-preference decision data comes from Become Specifically, for the first set of sales factor multi-preference decision data... A view First, we select sample points. Let's assume we have... The data matrix is , express The feature dimensions, using the principle of rationality from each Extracting intervals , will belong to The row index corresponding to the data inside is stored. By calculating the union of all dimensions ,exist Select the top ones with the highest frequency There are 1 index, denoted as _ ... ,Will The index in corresponds to The data in the middle constitutes a new For the selected Using the same method, representative features are selected from the representative data to obtain new data after selecting points and features. The corresponding data matrix is That is, to obtain multi-preference decision data on sales factors from Become .
[0011] The aforementioned principle of rationality specifically refers to: selecting sample points and features based on the principle of rationality, for... Each of them An interval needs to be defined. , and These are the lower and upper bounds of the selected representative data, respectively; that is, the representative data is defined by the interval... It is determined that the optimal value is obtained by maximizing the objective function. and The objective function is as follows:
[0012]
[0013] in, for The One data point, express The number of data points Representing one-dimensional data The median, function It is used to reflect the interval requirement, function It is used to reflect semantic requirements.
[0014] The process involves generating multiple view groups from each view of the selected sales factor multi-preference decision data through permutation and combination. The permutation and combination method is as follows: selecting... Combine the views to obtain The view group, the first A group of views is represented as: ; A group of views, represented as: ,in, The number of view cells in the view group. For the first The first view group Each view unit.
[0015] The process involves reconstructing data for each view unit in a view group to obtain multiple sparse sub-bipartite graphs, and then fusing these multiple sparse sub-bipartite graphs to obtain a joint view sparse bipartite graph. Specifically, this involves using a multi-view anchor point selection strategy based on view group collaboration to obtain anchor points for each view unit in a view group, thereby constructing a sparse sub-bipartite graph, and then fusing these multiple sparse sub-bipartite graphs to obtain a joint view sparse bipartite graph for the view group.
[0016] The process involves employing a multi-view anchor point selection strategy based on view group collaboration to obtain anchor points for view units within each view group, thereby constructing a sparse sub-bipartite graph. Specifically, for the first... The first view group View unit The data matrix is A multi-view anchor point selection strategy based on view group collaboration is adopted, prioritizing the selection of... All If the number of points shared by all views does not meet the requirement... Then continue selecting the view unit. All -1 unique common points of a view combination, and so on until the number of selected points is 1. By progressively expanding the selection range to meet the specified number of anchor points, the selected... Each point is denoted as a keypoint. The distance between each keypoint and all sample points in the data matrix is calculated. For each keypoint, the closest point in the data matrix is selected. For each sample point, the feature mean of the sample points is calculated as the anchor point. After obtaining the anchor point, the distance metric is applied only to each sample point. Reservation and Connect the nearest anchor points to construct a sparse sub-bipartite graph.
[0017] The view joint sparse bipartite graph is partitioned and integrated to obtain the final clustering result. Specifically, the spectral clustering algorithm is used to cluster each sparse bipartite graph in the view joint sparse bipartite graph to obtain the initial clustering result for each single view data. A similarity matrix is constructed based on the initial clustering result. The similarity matrices of all single view data are superimposed, and the average value of the superimposed matrix is calculated to obtain a comprehensive similarity matrix. The spectral clustering algorithm is then used again to cluster the comprehensive similarity matrix to obtain the final clustering result.
[0018] A sales factor multi-preference decision evaluation system based on data reconstruction and multi-view clustering integration includes a data processing module, a data reconstruction module, and a preference decision module; The data processing module is used to acquire multi-view data of sales factor multi-preference decision-making, and to arrange and combine each view data to generate multiple view groups; The data reconstruction module is used to reconstruct the data of the view units in each view group to obtain multiple sparse sub-bipartite graphs, and to merge the multiple sparse sub-bipartite graphs to obtain a view joint sparse bipartite graph. The preference decision module is used to partition and integrate the view joint sparse bipartite graph to obtain the final clustering result; and to evaluate sales factors based on the final clustering result.
[0019] Compared with the prior art, the present invention has the following beneficial effects: This invention arranges and combines multi-view data to generate multiple view groups, essentially cross-integrating sales data from different dimensions. The comprehensive utilization of multiple views can overcome the limitations of single data, improving the accuracy and robustness of data analysis. By using multi-view data for analysis, it breaks through the information limitations of a single view, comprehensively covering the correlations of sales factors in diverse scenarios, thus better meeting the complexity of consumer needs and laying the foundation for fully exploring data diversity. Simultaneously, this invention reconstructs the data of each view unit in each view group and generates sparse sub-bipartite graphs, which are then merged into a joint sparse bipartite graph. Essentially, it transforms scattered multi-view data into a unified structured relationship network. By mining the potential correlations between views, complex sales data is transformed into analyzable structured information, effectively capturing data complexity and avoiding the one-sidedness of single-view analysis. This invention partitions and integrates a joint sparse bipartite graph to obtain the final clustering result. It synthesizes common patterns from multiple views to filter out random errors and strengthens the prominence of key factors through an ensemble strategy. Ultimately, it achieves precise positioning of key sales factors, providing a reliable basis for sales decisions and solving the decision-making bias problem caused by incomplete information in traditional methods. By combining the prediction results of multiple basic models, this invention significantly improves the overall prediction accuracy and generalization ability. It also improves the training speed and efficiency of the model to a certain extent. Furthermore, it can typically process multiple basic models in parallel, thus accelerating model training and prediction, and performs excellently when handling large-scale data.
[0020] Specifically, this invention unifies the scale of all view data through standardized preprocessing, eliminating the interference of dimensional differences on subsequent analysis and ensuring the comparability of multi-view data. At the same time, it selects representative sample points and features based on the principle of rationality, and filters core data by defining intervals. While significantly reducing the amount of data, it fully preserves the core distribution characteristics and original main features of the data, avoids the interference of invalid information on the analysis results, and provides a high-quality data foundation for subsequent data reconstruction and clustering.
[0021] Furthermore, this invention generates multiple view groups by arranging and combining multi-view data after selecting points, thereby achieving cross-fusion of sales data from different dimensions. This breaks through the information limitations of a single view, comprehensively covers the correlation between sales factors in multiple scenarios, and thus better meets the complexity of consumers' actual needs, solving the problem of insufficient data diversity mining in traditional single-view analysis.
[0022] Furthermore, this invention employs an anchor point selection strategy based on view group collaboration, prioritizing common points among views and gradually expanding to neighboring views to ensure that anchor points reflect both common features and individual information. On this basis, a sparse sub-bipartite graph is constructed and merged, transforming scattered multi-view data into a unified structured relationship network. This effectively uncovers potential connections between views, avoids the one-sidedness of single-view analysis, and significantly improves the accuracy and reliability of data reconstruction.
[0023] Furthermore, after obtaining the initial clustering results of a single view through spectral clustering, this invention superimposes the similarity matrix of all views and calculates the mean, thus integrating the common patterns of multiple views to weaken the random error of a single view; then, it generates the final result through secondary clustering, further filtering out interfering information, and finally achieving accurate positioning of key sales factors. This solves the decision-making bias problem caused by incomplete information in traditional methods and provides reliable support for sales strategy formulation. Attached Figure Description
[0024] Figure 1 This is a schematic diagram of a sales factor multi-preference decision evaluation method based on data reconstruction and multi-view clustering integration in an embodiment of the present invention. Detailed Implementation
[0025] To further understand the content of this invention, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments are merely illustrative and not limiting of the invention.
[0026] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0027] This embodiment proposes a multi-view clustering ensemble method for evaluating sales factors based on data reconstruction and multi-preference decision-making, including the following steps: Acquire multi-preference decision data on sales factors, and arrange and combine the data from each view to generate multiple view groups; Data reconstruction is performed on the view units in each view group to obtain multiple sparse sub-bipartite graphs. These multiple sparse sub-bipartite graphs are then merged to obtain a joint sparse bipartite graph of the views. The view joint sparse bipartite graph is partitioned and integrated to obtain the final clustering result; based on the final clustering result, a multi-preference decision evaluation of sales factors is carried out.
[0028] To facilitate understanding of the embodiments of the present invention, this embodiment provides a detailed description of the sales factor multi-preference decision evaluation method based on data reconstruction and multi-view clustering ensemble, as follows: First, sales factor preference matrix data was obtained through a questionnaire survey. A survey on the impact of sales factors on sales volume was published on a website. There were three types of surveys, all 8×8 matrices: a probability preference relation matrix, a fuzzy preference relation matrix, and an interval preference relation matrix. The preference relation is used to represent the degree of preference between two objects. In this embodiment, a total of 2,000 consumers participated in the survey. For the probability preference matrix, we provide an interval [0,1]. Consumers fill in the table according to the interval and their own evaluation (for example, the probability of preference for price over appearance is 0.7 (0.6), where 0.7 means that the consumer thinks the price factor is more important than the appearance factor by 0.7, and 0.6 means that the probability of choosing this option is 0.6). For the fuzzy preference matrix, the given interval value is [0,1] (for example, the probability of preference for price over appearance is 0.7, that is, 0.7 means that the consumer thinks the price factor is more important than the appearance factor by 0.7). For the interval preference matrix, the given interval value is [0,1]. For example, the probability of preference for price over appearance is [0.2,0.5], which means that the consumer's preference level is an interval value of [0.2,0.5]. Similarly, the consumer thinks the price factor is more important than the appearance factor by [0.2,0.5].
[0029] The collected survey matrix data of the three types are processed as follows: First, the average value of each value in all the interval preference relation matrices is taken. For example, the average value of [0.2, 0.5] is 0.35. Then, the average value of the eight rows of elements in all preference relation matrices is taken, which gives the average relative importance of each consumer to the eight factors in the three different matrices. Finally, based on the above data and the three types of preference matrices, the data of each preference matrix is integrated to obtain an unlabeled multi-view data of sales factor multi-preference decision-making with three preference matrices. Each view data is a preference matrix, and each view data is a 2000×8 matrix. In this data, the first row contains eight sales factors, such as product, price, appearance, location, demand, marketing method, after-sales service, etc. The first column contains 2000 consumers, that is, each row of data represents one consumer's assessment of the importance of the eight factors.
[0030] The collected unlabeled sales factor multi-preference decision multi-view data were standardized using the following formula:
[0031] in, For initial sales factors, multiple preference decision data, To standardize the preprocessed sales factor preference decision data, All are known constants, and are generally taken as... That is, mapping data to .
[0032] After standardizing the collected unlabeled sales factor multi-preference decision multi-view data, sample points and features are selected for each view in turn. Data reconstruction is performed on the view units in each view group to obtain multiple sparse sub-bipartite graphs. These multiple sparse sub-bipartite graphs are then merged to obtain a joint view sparse bipartite graph. The specific method is as follows: First, representative sample points are selected sequentially for each view in the processed multi-preference sales factor decision data using a granular data selection method. Specifically, this process can be divided into two main stages: the first stage reduces the data volume while preserving the data's distribution characteristics through sample point selection; the second stage, based on the first stage, reduces the data dimensionality by dimensionality reduction while preserving the original key features of the data. That is, the multi-preference sales factor decision data is processed from... Become ,in This represents the first factor in the multi-preference decision dataset related to sales factors. One view, This represents the first of the multiple preference decision datasets for sales factors after point selection. One view, The total number of views for multi-preference decision-making data based on sales factors. The multi-preference decision-making data based on sales factors originates from... Become Specifically, for the first set of sales factor multi-preference decision data... A view First, we select sample points. Let's assume we have... The data matrix is , express The feature dimensions, using the principle of rationality from each Extracting intervals , will belong to The row index corresponding to the data inside is stored. By calculating the union of all dimensions ,exist Select the top ones with the highest frequency There are 1 index, denoted as _ ... ,Will The index in corresponds to The data in the middle constitutes a new For the selected Using the same method, representative features are selected from the representative data to obtain new data after selecting points and features. The corresponding data matrix is That is, to obtain multi-preference decision data on sales factors from Become .
[0033] Specifically, sales factors, preferences, decision-making, and multi-view data. ( (a view representation of X), for each view data ,in For sales factors, multiple preference decisions, and multi-view datasets, the first... Each view data, for Each column of data Sort the data in ascending order and find its median. Use the median to divide the data into left and right parts, satisfying the following conditions: ,in and All data is sorted in ascending order. Parameters The calculation is done by... and Obtained separately. For example, its corresponding parameters By comparing differences The objective function corresponding to the value This is achieved by, for each possible upper bound According to formula (4), we have:
[0034] like This indicates that there is only one data point on the right, which is used directly as the lower bound. Therefore, there is Find all average By normalizing the parameters, the parameters can be obtained. :
[0035] Indicates all The maximum value in.
[0036] Furthermore, the selection of sample points and features is based on the principle of rationality. Each of them An interval needs to be defined. , and These are the lower and upper bounds of the selected representative data, respectively; that is, the representative data is defined by the interval... It is determined that the optimal value is obtained by maximizing the objective function. and The objective function is as follows:
[0037]
[0038] in, for The One data point, express The number of data points Representing one-dimensional data The median, function It is used to reflect the interval requirement, function It is used to reflect semantic requirements. Based on the optimal... and Each of these can be obtained. The optimal interval .
[0039] Repeating the above steps eight times yields the index for each of the eight columns of data that satisfies its range. Therefore, each... Union of all dimensions Take the union of all the indices obtained above and sort them in descending order according to their frequency of occurrence. This is done by calculating the union of all dimensions. ,exist Select the top ones with the highest frequency There are 1 index, denoted as _ ... ,Will The index in corresponds to The data in the middle constitutes a new For the selected Using the same method, representative features are selected from the representative data to obtain new data after selecting points and features. The corresponding data matrix is That is, to obtain multi-preference decision data on sales factors from Become In this embodiment, in Select the top 20% of the most frequently occurring indices, denoted as . ,Will The index in corresponds to Data in For the selected representative data, repeat the above process three times to obtain the sample points selected for each view after data granulation (the number of sample points selected for each view is 400, 400, and 400 respectively). That is, the multi-view data is derived from... Become .
[0040] In summary, a new view group is derived. This includes three views, each containing 400, 400, and 400 sample points respectively, selected after data granulation. After the selection of candidate points is complete, in... Based on this, representative features were selected from the three view data using data granulation, with a target number of features. Obtain multi-view data ( (The number of sample points selected for each view are 2, 2, and 2 respectively), there are .
[0041] For each view of the sales factor preference decision data after point selection, i.e. Each view in the view is combined to create multiple view groups, and the combination method is to select the view group in sequence. Grouping views together, the number of view groups is... :
[0042] Each view of the multi-preference decision data for selected sales factors is combined to generate multiple view groups. The combination method is as follows: select... Combine the views to obtain The view group, the first A group of views is represented as: ; A group of views, represented as: ,in, The number of view cells in the view group. For the first The first view group Individual view units. In this embodiment, due to multiple sales factors and preferences in decision-making, multiple view data are used. Therefore, the number of view groups obtained is That is, Therefore, , , , .
[0043] according to For each view unit in a view group, a multi-view anchor point selection strategy based on view group collaboration is used to obtain anchor points, thereby constructing a sparse sub-bipartite graph. Multiple sparse sub-bipartite graphs are then merged to obtain a joint sparse bipartite graph of the view group. For the... The first view group View unit The data matrix is A multi-view anchor point selection strategy based on view group collaboration is adopted, prioritizing the selection of... All If the number of points shared by all views does not meet the requirement... Then continue selecting the view unit. All -1 unique common points of a view combination, and so on until the number of selected points is 1. By progressively expanding the selection range to meet the specified number of anchor points, the selected... Each point is denoted as a keypoint. The distance between each keypoint and all sample points in the data matrix is calculated. For each keypoint, the closest point in the data matrix is selected. For each sample point, the feature mean of the sample points is calculated as the anchor point. After obtaining the anchor point, the distance metric is applied only to each sample point. Reservation and Connect the nearest anchor points to construct a sparse sub-bipartite graph.
[0044] Specifically, firstly, based on the index set of the selected samples for each view. For each view group To obtain anchor points, due to the number of anchor points... Assuming ,Will The acquisition is evenly distributed among the view cells in the view group, so that... For example, each view unit number of anchor points .like ,because So for view groups View unit in The method for obtaining anchor points is as follows: calculate The common index of all views in the middle, i.e. The index points are counted, and it is determined whether the number of points satisfies 67. If not, the remaining points are counted from... Choose from a set of points, ensuring that no two points are selected repeatedly, and make selections based on this principle until only points are selected. Select from the options. Assign the selected point to... The sample data is denoted as keypoint data and is represented as follows: Find each key point separately. and The distance between all sample points in the keypoint is the distance between each keypoint. exist Select the 10 nearest sample points and calculate their feature mean as the anchor point representation: .
[0045] Secondly, based on the distance metric criterion, only for Each sample retains connections to its three nearest anchor points, based on the obtained anchor points. Constructing a sparse sub-bipartite graph The weight of its edge is and distance, Belongs to The three nearest anchor points, then There are 2000*3 non-zero elements in the array, and all other elements are 0. Therefore, we can obtain... Sparse sub-bipartite graph of other view cells By merging them, a view group is formed. View joint sparse bipartite graph The number of non-zero elements is 2000*3*3.
[0046] Furthermore, a spectral clustering algorithm is used to cluster each sparse bipartite graph in the view joint sparse bipartite graph to obtain the initial clustering result for each single view data. A similarity matrix is constructed based on the initial clustering result. The similarity matrices of all single view data are superimposed, and the average value of the superimposed matrix is calculated to obtain a comprehensive similarity matrix. The spectral clustering algorithm is used again to cluster the comprehensive similarity matrix to obtain the final clustering result. The sales factors are evaluated based on the final clustering result.
[0047] Specifically, for the view joint sparse bipartite graph These four views, combined with a sparse bipartite graph, are used for graph partitioning. Since... Similarity matrix Further Convert to a smaller image The initial clustering results for each single-view data were obtained using the spectral clustering algorithm. Based on the basic clustering results, similarity matrices are constructed as follows: A similarity matrix is a data structure used to represent the similarity between different objects. It is an important tool for analyzing data structures, performing pattern recognition, and optimizing algorithm performance. By quantifying the similarity between objects, it helps to reveal potential patterns and relationships in the data. It is a symmetric matrix, with all diagonal elements being 1, and the other elements representing the similarity between two samples. The formula for calculating similarity is: , Based on the above, we can conclude that The similarity matrix We perform a superposition to aggregate the information contained in the results of each basic clustering. Based on this, we calculate the average of the superimposed matrices to obtain a comprehensive similarity matrix. ,right Perform spectral clustering to generate the final clustering results, and evaluate sales factors.
[0048] Example 2 Based on the sales factor multi-preference decision evaluation method based on data reconstruction and multi-view clustering ensemble proposed in Example 1, this example proposes a sales factor multi-preference decision evaluation system based on data reconstruction and multi-view clustering ensemble, including a data processing module, a data reconstruction module, and a preference decision module; such as Figure 1 As shown, this embodiment achieves [the following] through a data processing module, a data reconstruction module, and a preference decision module. Figure 1 The illustrated methodology fully leverages the rich information in multi-view data, improving the accuracy and reliability of sales factor evaluation through data reconstruction and clustering integration, thus providing valuable references for sales decisions. Details are as follows: The data processing module is used to acquire unlabeled multi-view data of sales factor multi-preference decision-making and to preprocess the multi-view data of sales factor multi-preference decision-making.
[0049] The data reconstruction module is used to sequentially select sample points and features for each view group to generate a view group in a permutation and combination manner; to perform data reconstruction on the view units in each view group to obtain a sparse sub-bipartite graph, and to merge multiple sparse sub-bipartite graphs into a view joint sparse bipartite graph of the view group.
[0050] The preference decision module is used to obtain diverse basic clustering results based on the joint sparse bipartite graph of views obtained from multiple view groups, integrate them to obtain the final clustering result, and evaluate sales factors.
[0051] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that can be understood by those skilled in the art. The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.
Claims
1. A sales factor multi-preference decision evaluation method based on multi-view clustering ensemble of data reconstruction, characterized in that, Includes the following steps: Acquire multi-view data on sales factors, preferences, and decisions; arrange and combine each view data to generate multiple view groups. Data reconstruction is performed on the view units in each view group to obtain multiple sparse sub-bipartite graphs. These multiple sparse sub-bipartite graphs are then merged to obtain a joint sparse bipartite graph of the views. The view joint sparse bipartite graph is partitioned and integrated to obtain the final clustering result; Based on the final clustering results, sales factors are evaluated.
2. The sales factor multi-preference decision evaluation method based on data reconstruction and multi-view clustering integration as described in claim 1, characterized in that, The acquisition of multi-preference decision data based on sales factors involves permuting and combining data from each view to generate multiple view groups. Specifically, the acquired multi-preference decision data is preprocessed using standardization. Then, for each view in the processed multi-preference decision data, representative sample points are selected sequentially using a granular data selection method. This process can be divided into two main stages: the first stage reduces the data volume while preserving the data's distribution characteristics through sample point selection; the second stage, based on the first stage, reduces the data dimensionality by dimensionality reduction while preserving the original key features of the data. That is, the multi-preference decision data based on sales factors is transformed from... Become ,in This represents the first factor in the multi-preference decision dataset related to sales factors. One view, This represents the first of the multiple preference decision datasets for sales factors after point selection. One view, The total number of views for the multi-preference decision data of sales factors is determined; then, each view of the multi-preference decision data of sales factors after selection is used to generate multiple view groups through permutation and combination.
3. The sales factor multi-preference decision evaluation method based on data reconstruction and multi-view clustering integration according to claim 2, characterized in that, The acquired sales factor preference decision data is subjected to standardized preprocessing as follows: in, For initial sales factors, multiple preference decision data, To standardize the preprocessed sales factor preference decision data, All of these are known constants.
4. The sales factor multi-preference decision evaluation method based on data reconstruction and multi-view clustering integration according to claim 2, characterized in that, The sales factor multi-preference decision data comes from Become Specifically, for the first set of sales factor multi-preference decision data... A view First, we select sample points. Let's assume we have... The data matrix is , express The feature dimensions, using the principle of rationality from each Extracting intervals , will belong to The row index corresponding to the data inside is stored. By calculating the union of all dimensions ,exist Select the top ones with the highest frequency There are 1 index, denoted as _ ... ,Will The index in corresponds to The data in the middle constitutes a new For the selected Using the same method, representative features are selected from the representative data to obtain new data after selecting points and features. The corresponding data matrix is That is, to obtain multi-preference decision data on sales factors from Become .
5. The sales factor multi-preference decision evaluation method based on data reconstruction and multi-view clustering integration according to claim 3, characterized in that, The aforementioned principle of rationality specifically refers to: selecting sample points and features based on the principle of rationality, for... Each of them An interval needs to be defined. , and These are the lower and upper bounds of the selected representative data, respectively; that is, the representative data is defined by the interval... It is determined that the optimal value is obtained by maximizing the objective function. and The objective function is as follows: in, for The One data point, express The number of data points Representing one-dimensional data The median, function It is used to reflect the interval requirement, function It is used to reflect semantic requirements.
6. The sales factor multi-preference decision evaluation method based on data reconstruction and multi-view clustering integration according to claim 3, characterized in that, The process involves generating multiple view groups from each view of the selected sales factor multi-preference decision data through permutation and combination. The permutation and combination method is as follows: selecting... Combine the views to obtain The view group, the first A group of views is represented as: ; A group of views, represented as: ,in, The number of view cells in the view group. For the first The first view group Each view unit.
7. The sales factor multi-preference decision evaluation method based on data reconstruction and multi-view clustering integration according to claim 1, characterized in that, The process involves reconstructing data for each view unit in a view group to obtain multiple sparse sub-bipartite graphs, and then fusing these multiple sparse sub-bipartite graphs to obtain a joint view sparse bipartite graph. Specifically, this involves using a multi-view anchor point selection strategy based on view group collaboration to obtain anchor points for each view unit in a view group, thereby constructing a sparse sub-bipartite graph, and then fusing these multiple sparse sub-bipartite graphs to obtain a joint view sparse bipartite graph for the view group.
8. The sales factor multi-preference decision evaluation method based on data reconstruction and multi-view clustering integration according to claim 7, characterized in that, The process involves employing a multi-view anchor point selection strategy based on view group collaboration to obtain anchor points for view units within each view group, thereby constructing a sparse sub-bipartite graph. Specifically, for the first... The first view group View unit The data matrix is A multi-view anchor point selection strategy based on view group collaboration is adopted, prioritizing the selection of... All If the number of points shared by all views does not meet the requirement... Then continue selecting the view unit. All -1 unique common points of a view combination, and so on until the number of selected points is 1. By progressively expanding the selection range to meet the specified number of anchor points, the selected... Each point is denoted as a keypoint. The distance between each keypoint and all sample points in the data matrix is calculated. For each keypoint, the closest point in the data matrix is selected. For each sample point, the feature mean of the sample points is calculated as the anchor point. After obtaining the anchor point, the distance metric is applied only to each sample point. Reservation and Connect the nearest anchor points to construct a sparse sub-bipartite graph.
9. The sales factor multi-preference decision evaluation method based on data reconstruction and multi-view clustering integration according to claim 1, characterized in that, The view joint sparse bipartite graph is partitioned and integrated to obtain the final clustering result. Specifically, the spectral clustering algorithm is used to cluster each sparse bipartite graph in the view joint sparse bipartite graph to obtain the initial clustering result for each single view data. A similarity matrix is constructed based on the initial clustering result. The similarity matrices of all single view data are superimposed, and the average value of the superimposed matrix is calculated to obtain a comprehensive similarity matrix. The spectral clustering algorithm is then used again to cluster the comprehensive similarity matrix to obtain the final clustering result.
10. A sales factor multi-preference decision evaluation system based on multi-view clustering ensemble of data reconstruction, based on the sales factor multi-preference decision evaluation method based on multi-view clustering ensemble of data reconstruction as described in any one of claims 1 to 9, characterized in that, It includes a data processing module, a data reconstruction module, and a preference decision module; The data processing module is used to acquire multi-view data of sales factor multi-preference decision-making, and to arrange and combine each view data to generate multiple view groups; The data reconstruction module is used to reconstruct the data of the view units in each view group to obtain multiple sparse sub-bipartite graphs, and to merge the multiple sparse sub-bipartite graphs to obtain a view joint sparse bipartite graph. The preference decision module is used to partition and integrate the view joint sparse bipartite graph to obtain the final clustering result, and to evaluate sales factors based on the final clustering result.