A method, system and platform for recommending and optimizing enterprise collective purchasing plans

Abstract

The present invention discloses a method, system and platform for optimizing the recommendation of enterprise collective procurement plans. User procurement behavior big data is collected through terminals, and after feature analysis, similar procurement behavior data is integrated to form a historical procurement plan. Plan text data is generated based on the historical procurement plan, and the bag-of-words model is combined to perform text segmentation, unified vocabulary generation and text vector conversion, and a relationship table of object keywords is constructed by vector similarity comparison. Based on the relationship table, a knowledge graph is constructed in combination with procurement object attribute data, and rule reasoning and relationship expansion are performed through the AMIE+ algorithm. During the interaction cycle, user data is collected in real time, real-time object keywords and their semantic weights are analyzed, and a user behavior matrix and an associated behavior matrix are constructed. The effective associated matrix is screened out by similarity calculation between matrices, and the associated recommended procurement objects are parsed and marked to generate procurement recommendation information.

Description

Technical Field

[0001] The present invention relates to the field of procurement platform analysis, and more specifically, to a method, system and platform for recommending and optimizing enterprise collective procurement plans. Background Art

[0002] With the rapid development of e-commerce, more and more businesses are turning to e-commerce platforms for collective purchasing to improve efficiency and reduce costs. However, faced with a vast amount of product information and complex purchasing requirements, businesses or users often struggle to quickly find the most suitable procurement solution. Existing e-commerce platforms have numerous shortcomings in recommending procurement solutions, such as single-minded recommendation algorithms, a lack of personalization, insufficient predictive capabilities, and insufficient analysis of user behavior patterns, making them unable to meet the actual needs of businesses.

[0003] Therefore, there is an urgent need for an enterprise collective procurement plan recommendation optimization method to solve the above problems. Summary of the Invention

[0004] The present invention overcomes the defects of the prior art and proposes a method, system and platform for recommending and optimizing enterprise collective procurement plans.

[0005] A first aspect of the present invention provides a method for optimizing enterprise collective purchasing plans, comprising:

[0006] Analyze the behavior characteristics of the big data of purchasing behavior obtained by e-commerce purchasing terminals, integrate the purchasing plans corresponding to similar purchasing behaviors, and form a variety of historical purchasing plan information;

[0007] Format the historical procurement plan information and segment the procurement objects using a bag-of-words model to generate a unified vocabulary. The unified vocabulary includes multiple object keywords, each of which is associated with a procurement object.

[0008] By unifying the vocabulary, multiple planned text data are vectorized to obtain multiple text vectors;

[0009] Based on multiple text vectors, multiple groups of object keyword vectors are generated, and a relationship table is stored based on the similarity between the object keyword vectors;

[0010] Using object keywords as entity data, we set the relationship data between entity data based on the relationship table to build a knowledge graph, and then use the AMIE+ algorithm to perform rule reasoning and relationship data expansion on the knowledge graph.

[0011] During an interaction cycle, user purchasing behavior data is collected for feature analysis. Real-time object keywords are analyzed using a unified vocabulary. Entities are identified in the knowledge graph based on the real-time object keywords, and semantic weights are calculated based on the entity relationship states. A user behavior matrix is constructed using the semantic weights. Object keywords associated with the real-time object keywords are identified in the knowledge graph, and an associated behavior matrix is constructed.

[0012] The similarity calculation is performed between the user behavior matrix and multiple associated behavior matrices to screen out the effective associated matrix. Based on the effective associated matrix, the associated procurement objects are marked, and the procurement plan is set and recommended on the e-commerce procurement terminal to generate procurement recommendation information.

[0013] In this solution, the big data of purchasing behavior acquired by the e-commerce purchasing terminal is analyzed for behavioral characteristics, and the purchasing plans corresponding to similar purchasing behaviors are integrated to form a variety of historical purchasing plan information, specifically:

[0014] Obtaining user purchasing behavior big data through e-commerce purchasing terminals, wherein the purchasing behavior big data includes multiple historical user purchasing plans;

[0015] Conduct multi-dimensional behavioral semantic analysis based on all historical user purchase plans, and set multiple behavioral feature vectors based on semantic perspectives;

[0016] Based on the standard Euclidean distance, the similarity between behavioral feature vectors is calculated. If the similarity is within a preset range, they are classified into the same group, and finally multiple groups of behavioral feature vectors are generated;

[0017] Based on the division of behavioral feature vectors, historical user procurement plans are mapped and grouped, and historical user procurement plans in the same group are integrated to form a variety of historical procurement plan information.

[0018] In this solution, the historical procurement plan information is formatted and the procurement object is segmented using a bag-of-words model to generate a unified vocabulary. The unified vocabulary includes multiple object keywords, each of which is associated with a procurement object, specifically:

[0019] Perform text formatting on various historical procurement plan information to generate multiple plan text data, and perform text standardization preprocessing on the plan text data;

[0020] Based on the bag-of-words model, all planned text data are segmented, and keywords are extracted based on procurement objects. Unique words are counted to generate a unified vocabulary.

[0021] In this solution, the text vectorization is performed on multiple planned text data through a unified vocabulary to obtain multiple text vectors, specifically:

[0022] Performing text vectorization on multiple planned text data according to a unified vocabulary to form multiple text vectors;

[0023] Extract the vector representation of each object keyword from a text vector to generate a set of object keyword vectors;

[0024] The object keyword vector is stored in the system database.

[0025] In this solution, multiple groups of object keyword vectors are generated based on multiple text vectors, and a relationship table is stored based on the similarity between the object keyword vectors, specifically:

[0026] Generate multiple sets of object keyword vectors based on multiple text vectors;

[0027] Select two groups of object keyword vectors for vector similarity comparison and mark them as the first vector group and the second vector group;

[0028] Selecting an object keyword vector from each of the first vector group and the second vector group to perform similarity calculation, the similarity calculation is based on the Mahalanobis distance calculation of the vectors, and setting the strength relationship between the two object keywords based on the similarity results;

[0029] In the absence of repetition, two object keyword vectors are cyclically selected for similarity calculation, and a relationship table between the object keywords is generated;

[0030] Two groups of object keyword vectors are cyclically selected for calculation, and the relationship table is updated.

[0031] In this solution, the object keywords are used as entity data, and the relationship data between the entity data is set based on the relationship table to construct a knowledge graph. The AMIE+ algorithm is used to perform rule reasoning and relationship data expansion on the knowledge graph. Specifically:

[0032] In the unified vocabulary, set the corresponding entity data with each object keyword;

[0033] From the big data of purchasing behavior, obtain the purchasing object attribute data corresponding to the object keyword and use it as the entity attribute data;

[0034] Setting the relationship data between the entity data based on the relationship table and constructing a knowledge graph of the procurement object;

[0035] Based on the existing relationship model of the knowledge graph, a rule queue is set up, and the rule queue is combined with the AMIE+ algorithm to perform rule reasoning and expansion to generate association rules;

[0036] Expand relational data based on association rules and update the knowledge graph.

[0037] In this solution, within an interaction cycle, user purchasing behavior data is collected for feature analysis. Real-time object keywords are analyzed in combination with a unified vocabulary. Corresponding entities are identified in the knowledge graph based on the real-time object keywords, and semantic weights are calculated based on the entity relationship states. A user behavior matrix is constructed based on the semantic weights. Object keywords associated with the real-time object keywords are identified in the knowledge graph and an associated behavior matrix is constructed. Specifically,

[0038] During an interaction cycle, user purchasing behavior data is collected in real time;

[0039] The purchasing behavior data is formatted and the semantic analysis model is used to perform semantic analysis and word segmentation. A unified vocabulary is introduced for keyword statistics to obtain multiple real-time object keywords.

[0040] Perform entity retrieval and identification in the knowledge graph based on real-time object keywords to obtain multiple labeled entities;

[0041] Based on a tagged entity, in the knowledge graph, count the number of associated entities N of the tagged entity. For each N associated entity, evaluate the association strength between the N associated entities and the tagged entity in combination with the relationship table, and calculate the average association strength. Finally, calculate the average association strength P of each identified entity.

[0042] A semantic weight is obtained by performing a weighted average of N and P, multiple semantic weights are calculated based on multiple tagged entities, and the multiple semantic weights are associated with multiple real-time object keywords;

[0043] Through a unified vocabulary, we obtain the index values of multiple real-time object keywords, and construct a user behavior matrix with the index value as the first dimension and the semantic weight as the second dimension.

[0044] In the knowledge graph, object keywords associated with real-time object keywords are extracted and grouped. The extraction process is to obtain multiple related entities associated with the identification entity, group the multiple related entities based on the strength of the association, so that the related entities in the same group have the same strength of association with the identification entity, and map them to the associated object keywords based on the grouping situation to obtain multiple groups of related keywords;

[0045] Based on each group of related keywords, the corresponding multiple related keywords are marked as entities in the knowledge graph and the corresponding behavior matrix is calculated and marked as the related behavior matrix;

[0046] Multiple association behavior matrices are formed based on multiple groups of associated keywords.

[0047] In this solution, similarity calculation is performed between the user behavior matrix and multiple association behavior matrices to screen out the effective association matrix, and the related purchase objects are marked based on the effective association matrix. The purchase plan is set and recommended at the e-commerce purchase terminal to generate purchase recommendation information. Specifically:

[0048] The Jordan standard similarity determination method is introduced to calculate the similarity between the user behavior matrix and multiple related behavior matrices;

[0049] Filter out the correlation behavior matrices whose similarity is greater than a preset threshold to obtain a valid correlation matrix;

[0050] Perform data analysis on the effective association matrix and mark the associated procurement objects;

[0051] Based on the associated procurement objects, the e-commerce procurement terminal conducts procurement object attribute analysis and supplier supply status correlation analysis, performs procurement plan setting and recommendation at the e-commerce procurement terminal, and generates procurement recommendation information.

[0052] A second aspect of the present invention further provides a system for recommending and optimizing enterprise collective procurement plans. The system includes a memory and a processor. The memory includes an enterprise collective procurement plan recommendation and optimization program. When the enterprise collective procurement plan recommendation and optimization program is executed by the processor, the following steps are implemented:

[0053] Analyze the behavior characteristics of the big data of purchasing behavior obtained by e-commerce purchasing terminals, integrate the purchasing plans corresponding to similar purchasing behaviors, and form a variety of historical purchasing plan information;

[0054] Format the historical procurement plan information and segment the procurement objects using a bag-of-words model to generate a unified vocabulary. The unified vocabulary includes multiple object keywords, each of which is associated with a procurement object.

[0055] By unifying the vocabulary, multiple planned text data are vectorized to obtain multiple text vectors;

[0056] Based on multiple text vectors, multiple groups of object keyword vectors are generated, and a relationship table is stored based on the similarity between the object keyword vectors;

[0057] Using object keywords as entity data, we set the relationship data between entity data based on the relationship table to build a knowledge graph, and then use the AMIE+ algorithm to perform rule reasoning and relationship data expansion on the knowledge graph.

[0058] During an interaction cycle, user purchasing behavior data is collected for feature analysis. Real-time object keywords are analyzed using a unified vocabulary. Entities are identified in the knowledge graph based on the real-time object keywords, and semantic weights are calculated based on the entity relationship states. A user behavior matrix is constructed using the semantic weights. Object keywords associated with the real-time object keywords are identified in the knowledge graph, and an associated behavior matrix is constructed.

[0059] The similarity calculation is performed between the user behavior matrix and multiple associated behavior matrices to screen out the effective associated matrix. Based on the effective associated matrix, the associated procurement objects are marked, and the procurement plan is set and recommended on the e-commerce procurement terminal to generate procurement recommendation information.

[0060] The third aspect of the present invention also provides an enterprise collective procurement plan recommendation optimization platform, which includes: a data acquisition module, a data processing module, and a visualization module. When the platform is running, it implements the steps of the enterprise collective procurement plan recommendation optimization method as described in any of the above items.

[0061] The present invention discloses a method, system and platform for optimizing the recommendation of enterprise collective procurement plans. User procurement behavior big data is collected through terminals, and after feature analysis, similar procurement behavior data is integrated to form a historical procurement plan. Plan text data is generated based on the historical procurement plan, and the bag-of-words model is combined to perform text segmentation, unified vocabulary generation and text vector conversion, and a relationship table of object keywords is constructed by vector similarity comparison. Based on the relationship table, a knowledge graph is constructed in combination with procurement object attribute data, and rule reasoning and relationship expansion are performed through the AMIE+ algorithm. During the interaction cycle, user data is collected in real time, real-time object keywords and their semantic weights are analyzed, and a user behavior matrix and an associated behavior matrix are constructed. The effective associated matrix is screened out by similarity calculation between matrices, and the associated recommended procurement objects are parsed and marked to generate procurement recommendation information. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 A flow chart showing a method for recommending and optimizing enterprise collective purchasing plans according to the present invention is shown;

[0063] Figure 2 A block diagram of a system for recommending and optimizing enterprise collective purchasing plans according to the present invention is shown. DETAILED DESCRIPTION

[0064] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0065] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0066] Figure 1 A flow chart of an enterprise collective purchasing plan recommendation optimization method of the present invention is shown.

[0067] like Figure 1 As shown, the first aspect of the present invention provides a method for optimizing enterprise collective purchasing plans, which specifically includes the following steps:

[0068] S102: Analyze the behavior characteristics of the purchasing behavior big data obtained by the e-commerce purchasing terminal, merge the purchasing plans corresponding to similar purchasing behaviors, and form multiple historical purchasing plan information;

[0069] S104: Format the historical procurement plan information and perform word segmentation on the procurement objects using a bag-of-words model to generate a unified vocabulary. The unified vocabulary includes multiple object keywords, each of which is associated with a procurement object.

[0070] S106, performing text vectorization on the plurality of planned text data using a unified vocabulary to obtain a plurality of text vectors;

[0071] S108, generating multiple groups of object keyword vectors based on the multiple text vectors, and storing a relationship table based on similarities between the object keyword vectors;

[0072] S110, using object keywords as entity data, setting relationship data between entity data based on the relationship table, constructing a knowledge graph, and performing rule reasoning and relationship data expansion on the knowledge graph using the AMIE+ algorithm;

[0073] S112: During an interaction cycle, user purchasing behavior data is collected for feature analysis. Real-time object keywords are analyzed using a unified vocabulary. Entities corresponding to the real-time object keywords are identified in the knowledge graph, and semantic weights are calculated based on the entity relationship states. A user behavior matrix is constructed using the semantic weights. Object keywords associated with the real-time object keywords are identified in the knowledge graph, and an associated behavior matrix is constructed.

[0074] S114, similarity calculation is performed between the user behavior matrix and multiple associated behavior matrices to screen out valid associated matrices, related purchase objects are marked based on the valid associated matrices, and purchase plans are set and recommended at the e-commerce purchase terminal to generate purchase recommendation information.

[0075] According to an embodiment of the present invention, the S102 is specifically:

[0076] Obtaining user purchasing behavior big data through e-commerce purchasing terminals, wherein the purchasing behavior big data includes multiple historical user purchasing plans;

[0077] Conduct multi-dimensional behavioral semantic analysis based on all historical user purchase plans, and set multiple behavioral feature vectors based on semantic perspectives;

[0078] Based on the standard Euclidean distance, the similarity between behavioral feature vectors is calculated. If the similarity is within a preset range, they are classified into the same group, and finally multiple groups of behavioral feature vectors are generated;

[0079] Based on the division of behavioral feature vectors, historical user procurement plans are mapped and grouped, and historical user procurement plans in the same group are integrated to form a variety of historical procurement plan information.

[0080] It should be noted that the user's procurement plan includes information such as the quantity of items purchased by the user, the type of items, the procurement supplier, and the procurement time.

[0081] The multi-dimensional behavior semantic analysis is specifically performed based on a preset semantic analysis model, converting semantic information from data of different dimensions and generating multi-dimensional behavior feature vectors. Each user purchase plan includes a behavior feature vector. Each historical purchase plan information includes at least one user purchase plan.

[0082] The multi-dimensional behavior feature vector corresponds to multiple dimensions such as the quantity of purchased items, item type, purchase supplier, purchase time information, etc., and the value of each dimension is the corresponding purchase information value.

[0083] According to an embodiment of the present invention, the S104 is specifically as follows:

[0084] Perform text formatting on various historical procurement plan information to generate multiple plan text data, and perform text standardization preprocessing on the plan text data;

[0085] Based on the bag-of-words model, all planned text data are segmented, and keywords are extracted based on procurement objects. Unique words are counted to generate a unified vocabulary.

[0086] It should be noted that when counting unique words, duplicate words are removed, and different types of words are counted. Only one of each word is retained to generate a vocabulary. The unified vocabulary includes multiple object keywords. Object keywords are keywords for purchase objects and are the data representation of purchase objects in the semantic analysis model (or bag-of-words model). Purchase objects are purchased items, and there is a one-to-one correspondence between the two. Each object keyword is associated with a purchase object.

[0087] Each historical procurement plan includes multiple procurement objects. Specifically, the text vectors are the sum of the keyword vectors for these objects. Therefore, we need to extract keyword vectors from the text vectors and analyze each keyword. Analyzing each text vector corresponds to analyzing the procurement objects for each acquisition plan.

[0088] According to an embodiment of the present invention, S106 is specifically:

[0089] Performing text vectorization on multiple planned text data according to a unified vocabulary to form multiple text vectors;

[0090] Extract the vector representation of each object keyword from a text vector to generate a set of object keyword vectors;

[0091] The object keyword vector is stored in the system database.

[0092] It should be noted that the group of object keyword vectors includes at least one object keyword vector.

[0093] According to an embodiment of the present invention, the S108 is specifically as follows:

[0094] Generate multiple sets of object keyword vectors based on multiple text vectors;

[0095] Select two groups of object keyword vectors for vector similarity comparison and mark them as the first vector group and the second vector group;

[0096] Selecting an object keyword vector from each of the first vector group and the second vector group to perform similarity calculation, the similarity calculation is based on the Mahalanobis distance calculation of the vectors, and setting the strength relationship between the two object keywords based on the similarity results;

[0097] In the absence of repetition, two object keyword vectors are cyclically selected for similarity calculation, and a relationship table between the object keywords is generated;

[0098] Two groups of object keyword vectors are cyclically selected for calculation, and the relationship table is updated.

[0099] It should be noted that, in the setting of the strength relationship between the two object keywords based on the similarity results, the greater the calculated vector distance, the smaller the similarity, and the weaker the relationship between the object keywords. In the non-repetitive case, two object keyword vectors are cyclically selected for similarity calculation, and two groups of vectors are selected and calculated respectively. For example, there are two vectors in the first vector group and the second vector group respectively, marked as the first group (A, B vectors) and the second group (C, D vectors). The number of calculations is four times, corresponding to the similarity calculations between AC, AD, BC, and BD respectively. The two groups of object keyword vectors are cyclically selected for calculation in the same non-repetitive selection method. It can be understood here that the same object keyword vector may exist in two object keyword vector groups. Therefore, there are cases where two word vectors are calculated multiple times. In this case, the average distance value is used for relationship strength analysis.

[0100] Each vector group represents a procurement plan and also reflects a user procurement behavior pattern. When conducting relationship analysis, semantic analysis and similarity analysis in the form of word vectors based on the procurement items corresponding to different behavior patterns can explore the relationship between procurement objects in different procurement models. By recording the relationship, the corresponding knowledge graph is subsequently constructed, and the association between different procurement models and procurement items is effectively explored, thereby analyzing effective recommendation plans and information.

[0101] According to an embodiment of the present invention, the S110 is specifically:

[0102] In the unified vocabulary, set the corresponding entity data with each object keyword;

[0103] From the big data of purchasing behavior, obtain the purchasing object attribute data corresponding to the object keyword and use it as the entity attribute data;

[0104] Setting the relationship data between the entity data based on the relationship table and constructing a knowledge graph of the procurement object;

[0105] Based on the existing relationship model of the knowledge graph, a rule queue is set up, and the rule queue is combined with the AMIE+ algorithm to perform rule reasoning and expansion to generate association rules;

[0106] Expand relational data based on association rules and update the knowledge graph.

[0107] It should be noted that the entities in the knowledge graph are corresponding entity nodes or entity data, and the relationship data between entities are initialized through a relationship table. In the present invention, the knowledge graph can save the relationship and knowledge data between procurement objects based on the semantic level, which is used to mine the existing association relationships and procurement behavior feature relationships, and realize data recommendation more efficiently.

[0108] According to an embodiment of the present invention, the S112 is specifically as follows:

[0109] During an interaction cycle, user purchasing behavior data is collected in real time;

[0110] The purchasing behavior data is formatted and the semantic analysis model is used to perform semantic analysis and word segmentation. A unified vocabulary is introduced for keyword statistics to obtain multiple real-time object keywords.

[0111] Perform entity retrieval and identification in the knowledge graph based on real-time object keywords to obtain multiple labeled entities;

[0112] Based on a tagged entity, in the knowledge graph, count the number of associated entities N of the tagged entity. For each N associated entity, evaluate the association strength between the N associated entities and the tagged entity in combination with the relationship table, and calculate the average association strength. Finally, calculate the average association strength P of each identified entity.

[0113] A semantic weight is obtained by performing a weighted average of N and P, multiple semantic weights are calculated based on multiple tagged entities, and the multiple semantic weights are associated with multiple real-time object keywords;

[0114] Through a unified vocabulary, we obtain the index values of multiple real-time object keywords, and construct a user behavior matrix with the index value as the first dimension and the semantic weight as the second dimension.

[0115] In the knowledge graph, object keywords associated with real-time object keywords are extracted and grouped. The extraction process is to obtain multiple related entities associated with the identification entity, group the multiple related entities based on the strength of the association, so that the related entities in the same group have the same strength of association with the identification entity, and map them to the associated object keywords based on the grouping situation to obtain multiple groups of related keywords;

[0116] Based on each group of related keywords, the corresponding multiple related keywords are marked as entities in the knowledge graph and the corresponding behavior matrix is calculated and marked as the related behavior matrix;

[0117] Multiple association behavior matrices are formed based on multiple groups of associated keywords.

[0118] It should be noted that the association strength P is based on the average association strength between the corresponding entity node and the other associated nodes, and the association strength is obtained through the relationship table.

[0119] The semantic weight Q is calculated as follows:

[0120] Q=K1×N+K2×P;

[0121] Among them, K1 and K2 are preset weights, N is the number of associated entities, and P is the association strength.

[0122] A real-time object keyword corresponds to an identification entity and also corresponds to a semantic weight.

[0123] The user behavior matrix is specifically a data representation that reflects the characteristics of user purchasing behavior and the corresponding purchasing objects. By analyzing the number of associated objects and the relationship strength of the purchasing objects corresponding to the user's purchasing behavior in a certain interaction cycle, the matrix is constructed. This can effectively analyze the relationship weight of the user behavior pattern in the knowledge graph, and then construct a matrix with user behavior characteristics. It can describe the behavioral characteristics of user purchases in the semantic dimension, and through analysis of the knowledge graph model, it can realize efficient correlation data mining of behavioral characteristics, providing data support for efficient data recommendation.

[0124] In the extraction and grouping of object keywords associated with real-time object keywords, in the knowledge graph, real-time object keywords correspond to identification entities, and associated object keywords are associated entities. In addition, in the grouping process, multiple groups can be set, for example, three groups are set, and the corresponding relationship strengths are divided into three levels, one, two, and three. The relationship strengths between the associated entities and the identification entities corresponding to all associated keywords in the first group are all one level, that is, a stronger degree of relationship, and so on.

[0125] Based on each group of associated keywords, the corresponding multiple associated keywords are entity marked in the knowledge graph and the corresponding behavior matrix is calculated. The process is consistent with the process of constructing the matrix of the above-mentioned multiple marked entities, that is, the corresponding multiple associated keywords are used as multiple identification entities for matrix analysis and calculation.

[0126] In the process of analyzing the user behavior feature matrix and the associated behavior matrix, the relational model of the knowledge graph can effectively analyze the user's potential behavior patterns and mine the associated purchasing features and associated purchasing objects, and perform associated comparisons in the form of a matrix to achieve efficient recommendation data screening and obtain associated recommendation data suitable for different users.

[0127] According to an embodiment of the present invention, the S114 is specifically as follows:

[0128] The Jordan standard similarity determination method is introduced to calculate the similarity between the user behavior matrix and multiple related behavior matrices;

[0129] Filter out the correlation behavior matrices whose similarity is greater than a preset threshold to obtain a valid correlation matrix;

[0130] Perform data analysis on the effective association matrix and mark the associated procurement objects;

[0131] Based on the associated procurement objects, the e-commerce procurement terminal conducts procurement object attribute analysis and supplier supply status correlation analysis, performs procurement plan setting and recommendation at the e-commerce procurement terminal, and generates procurement recommendation information.

[0132] It should be noted that the procurement object attribute analysis refers to the analysis of item type, quantity, specifications, etc.

[0133] According to an embodiment of the present invention, the S112 further includes:

[0134] Set M consecutive interaction cycles;

[0135] Among M consecutive interaction cycles, a current interaction cycle is selected for analysis, and multiple related entities are grouped based on the strength of the association. During the grouping process, two groups of related keywords are set, the first group corresponds to strong associations, and the second group corresponds to weak associations;

[0136] Based on the semantic weights corresponding to the two groups of related keywords, a first related behavior matrix and a second related behavior matrix are constructed respectively;

[0137] Using the Jordan standard similarity determination method, the similarity between the first association behavior matrix and the user behavior matrix is calculated to obtain a first similarity, and the similarity between the second association behavior matrix and the user behavior matrix is calculated to obtain a second similarity;

[0138] Calculating a growth rate between a first similarity in a current interaction period and a first similarity in a next interaction period to obtain a first growth rate;

[0139] Calculating a growth rate between the second similarity of the current interaction period and the second similarity of the next interaction period to obtain a second growth rate;

[0140] If both the first growth rate and the second growth rate are positive, the current interaction period is set as the recommended period, otherwise it is set as the non-recommended period;

[0141] Analyze M consecutive interaction cycles, and generate a recommended time interval table for the M consecutive interaction cycles based on the setting process of recommended cycles and non-recommended cycles;

[0142] Apply the recommended time interval table to the next M consecutive interaction cycles;

[0143] Set different recommended time interval tables based on different users.

[0144] It should be noted that M is the number of preset cycles. Among M consecutive interaction cycles, the Mth (i.e., the last) interaction cycle may not be subjected to recommendation cycle analysis. A positive growth rate represents an increase in similarity, while a negative value represents a decrease in similarity. The present invention can effectively improve the recommendation efficiency by setting the recommended time node interval, and different recommendation time interval tables can be set based on different users, which are suitable for recommendation time nodes of different users. In the next M consecutive interaction cycles, data recommendations for the corresponding periodic time period are performed based on the set recommendation time interval table. On the one hand, it can effectively improve the recommendation efficiency and reduce redundant data analysis. On the other hand, it can also improve the rationality of the recommendation time.

[0145] Figure 2 A block diagram of a system for recommending and optimizing enterprise collective purchasing plans according to the present invention is shown.

[0146] A second aspect of the present invention further provides a system 2 for recommending and optimizing enterprise collective procurement plans. The system comprises: a memory 21 and a processor 22. The memory 21 includes an enterprise collective procurement plan recommendation and optimization program. When the enterprise collective procurement plan recommendation and optimization program is executed by the processor 22, the following steps are implemented:

[0147] S102: Analyze the behavior characteristics of the purchasing behavior big data obtained by the e-commerce purchasing terminal, merge the purchasing plans corresponding to similar purchasing behaviors, and form multiple historical purchasing plan information;

[0148] S104: Format the historical procurement plan information and perform word segmentation on the procurement objects using a bag-of-words model to generate a unified vocabulary. The unified vocabulary includes multiple object keywords, each of which is associated with a procurement object.

[0149] S106, performing text vectorization on the plurality of planned text data using a unified vocabulary to obtain a plurality of text vectors;

[0150] S108, generating multiple groups of object keyword vectors based on the multiple text vectors, and storing a relationship table based on similarities between the object keyword vectors;

[0151] S110, using object keywords as entity data, setting relationship data between entity data based on the relationship table, constructing a knowledge graph, and performing rule reasoning and relationship data expansion on the knowledge graph using the AMIE+ algorithm;

[0152] S112: During an interaction cycle, user purchasing behavior data is collected for feature analysis. Real-time object keywords are analyzed using a unified vocabulary. Entities corresponding to the real-time object keywords are identified in the knowledge graph, and semantic weights are calculated based on the entity relationship states. A user behavior matrix is constructed using the semantic weights. Object keywords associated with the real-time object keywords are identified in the knowledge graph, and an associated behavior matrix is constructed.

[0153] S114, similarity calculation is performed between the user behavior matrix and multiple associated behavior matrices to screen out valid associated matrices, related purchase objects are marked based on the valid associated matrices, and purchase plans are set and recommended at the e-commerce purchase terminal to generate purchase recommendation information.

[0154] Based on the actual operation process of the system, the system can implement the steps of the enterprise collective procurement plan recommendation optimization method as described in any of the above items.

[0155] The third aspect of the present invention also provides an enterprise collective procurement plan recommendation optimization platform, which includes: a data acquisition module, a data processing module, and a visualization module. When the platform is running, it implements the steps of the enterprise collective procurement plan recommendation optimization method as described in any of the above items.

[0156] The modules are:

[0157] The data acquisition module is used to collect data in real time.

[0158] The data processing module is used to process and analyze data, output processing results and save data.

[0159] The visualization module is used to visualize data and recommend data to users, thereby enabling data interaction with users.

[0160] The present invention discloses a method, system and platform for optimizing the recommendation of enterprise collective procurement plans. User procurement behavior big data is collected through terminals, and after feature analysis, similar procurement behavior data is integrated to form a historical procurement plan. Plan text data is generated based on the historical procurement plan, and the bag-of-words model is combined to perform text segmentation, unified vocabulary generation and text vector conversion, and a relationship table of object keywords is constructed by vector similarity comparison. Based on the relationship table, a knowledge graph is constructed in combination with procurement object attribute data, and rule reasoning and relationship expansion are performed through the AMIE+ algorithm. During the interaction cycle, user data is collected in real time, real-time object keywords and their semantic weights are analyzed, and a user behavior matrix and an associated behavior matrix are constructed. The effective associated matrix is screened out by similarity calculation between matrices, and the associated recommended procurement objects are parsed and marked to generate procurement recommendation information.

[0161] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0162] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0163] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0164] Those skilled in the art will appreciate that all or part of the steps of the above-mentioned method embodiments may be implemented by hardware associated with program instructions, and the aforementioned program may be stored in a computer-readable storage medium. When the program is executed, the program executes the steps of the above-mentioned method embodiments. The aforementioned storage medium includes various media that can store program codes, such as mobile storage devices, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0165] Alternatively, if the integrated units described above are implemented as software modules and sold or used as standalone products, they can also be stored on a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product, stored on a storage medium, includes instructions for enabling a computer device (such as a personal computer, server, or network device) to execute all or part of the methods described in various embodiments of the present invention. The aforementioned storage media include various media capable of storing program code, such as removable storage devices, ROM, RAM, magnetic disks, or optical disks.

[0166] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A method for optimizing enterprise collective purchasing plans, characterized in that: include: Analyze the behavior characteristics of the big data of purchasing behavior obtained by e-commerce purchasing terminals, integrate the purchasing plans corresponding to similar purchasing behaviors, and form a variety of historical purchasing plan information; Format the historical procurement plan information and segment the procurement objects using a bag-of-words model to generate a unified vocabulary. The unified vocabulary includes multiple object keywords, each of which is associated with a procurement object. By unifying the vocabulary, multiple planned text data are vectorized to obtain multiple text vectors; Based on multiple text vectors, multiple groups of object keyword vectors are generated, and a relationship table is stored based on the similarity between the object keyword vectors; Using object keywords as entity data, we set the relationship data between entity data based on the relationship table to build a knowledge graph, and then use the AMIE+ algorithm to perform rule reasoning and relationship data expansion on the knowledge graph. During an interaction cycle, user purchasing behavior data is collected for feature analysis. Real-time object keywords are analyzed using a unified vocabulary. Entities are identified in the knowledge graph based on the real-time object keywords, and semantic weights are calculated based on the entity relationship states. A user behavior matrix is constructed using the semantic weights. Object keywords associated with the real-time object keywords are identified in the knowledge graph, and an associated behavior matrix is constructed. Calculate the similarity between the user behavior matrix and multiple related behavior matrices and select the effective related matrix. Based on the effective related matrix, mark the related purchase objects, set and recommend purchase plans on the e-commerce purchase terminal, and generate purchase recommendation information. The user behavior matrix and the associated behavior matrix specifically include: During an interaction cycle, user purchasing behavior data is collected in real time; The purchasing behavior data is formatted and the semantic analysis model is used to perform semantic analysis and word segmentation. A unified vocabulary is introduced for keyword statistics to obtain multiple real-time object keywords. Entity retrieval and identification are performed in the knowledge graph based on real-time object keywords to obtain multiple labeled entities; Based on a tagged entity, in the knowledge graph, count the number of associated entities N of the tagged entity. For each N associated entity, evaluate the association strength between the N associated entities and the tagged entity in combination with the relationship table, and calculate the average association strength. Finally, calculate the average association strength P of each identified entity. A semantic weight is obtained by performing a weighted average of N and P, multiple semantic weights are calculated based on multiple tagged entities, and the multiple semantic weights are associated with multiple real-time object keywords; Through a unified vocabulary, we obtain the index values of multiple real-time object keywords, and construct a user behavior matrix with the index value as the first dimension and the semantic weight as the second dimension. In the knowledge graph, object keywords associated with real-time object keywords are extracted and grouped. The extraction process is to obtain multiple related entities associated with the identification entity, group the multiple related entities based on the strength of the association, so that the related entities in the same group have the same strength of association with the identification entity, and map them to the associated object keywords based on the grouping situation to obtain multiple groups of related keywords; Based on each group of related keywords, the corresponding multiple related keywords are marked as entities in the knowledge graph and the corresponding behavior matrix is calculated and marked as the related behavior matrix; Multiple association behavior matrices are formed based on multiple groups of associated keywords.

2. The enterprise collective procurement plan recommendation optimization method according to claim 1, characterized in that: The aforementioned method involves analyzing the behavior characteristics of the purchasing behavior big data obtained by the e-commerce purchasing terminal, integrating the purchasing plans corresponding to similar purchasing behaviors, and forming a variety of historical purchasing plan information, specifically: Obtaining user purchasing behavior big data through e-commerce purchasing terminals, wherein the purchasing behavior big data includes multiple historical user purchasing plans; Conduct multi-dimensional behavioral semantic analysis based on all historical user purchase plans, and set multiple behavioral feature vectors based on semantic perspectives; Based on the standard Euclidean distance, the similarity between behavioral feature vectors is calculated. If the similarity is within a preset range, they are classified into the same group, and finally multiple groups of behavioral feature vectors are generated; Based on the division of behavioral feature vectors, historical user procurement plans are mapped and grouped, and historical user procurement plans in the same group are integrated to form a variety of historical procurement plan information.

3. The enterprise collective procurement plan recommendation optimization method according to claim 1, characterized in that: The text formatting of the historical procurement plan information is performed and a word segmentation operation of the procurement object is performed using a bag-of-words model to generate a unified vocabulary. The unified vocabulary includes multiple object keywords, and each object keyword is associated with a procurement object, specifically: Perform text formatting on various historical procurement plan information to generate multiple plan text data, and perform text standardization preprocessing on the plan text data; Based on the bag-of-words model, all planned text data are segmented, and keywords are extracted based on procurement objects. Unique words are counted to generate a unified vocabulary.

4. The enterprise collective procurement plan recommendation optimization method according to claim 1, characterized in that: The text vectorization is performed on multiple planned text data by unifying the vocabulary to obtain multiple text vectors, specifically: Performing text vectorization on multiple planned text data according to a unified vocabulary to form multiple text vectors; Extract the vector representation of each object keyword from a text vector to generate a set of object keyword vectors; The object keyword vector is stored in the system database.

5. The enterprise collective procurement plan recommendation optimization method according to claim 1, characterized in that: The method generates multiple sets of object keyword vectors based on multiple text vectors, and stores a relationship table based on the similarity between the object keyword vectors, specifically: Generate multiple sets of object keyword vectors based on multiple text vectors; Select two groups of object keyword vectors for vector similarity comparison and mark them as the first vector group and the second vector group; Selecting an object keyword vector from each of the first vector group and the second vector group to perform similarity calculation, the similarity calculation is based on the Mahalanobis distance calculation of the vectors, and setting the strength relationship between the two object keywords based on the similarity results; In the absence of repetition, two object keyword vectors are cyclically selected for similarity calculation, and a relationship table between the object keywords is generated; Two groups of object keyword vectors are cyclically selected for calculation, and the relationship table is updated.

6. The enterprise collective procurement plan recommendation optimization method according to claim 5, characterized in that: The method uses object keywords as entity data, sets the relationship data between entity data based on the relationship table, constructs a knowledge graph, and uses the AMIE+ algorithm to perform rule reasoning and relationship data expansion on the knowledge graph. Specifically: In the unified vocabulary, set the corresponding entity data with each object keyword; From the big data of purchasing behavior, obtain the purchasing object attribute data corresponding to the object keyword and use it as the entity attribute data; Setting the relationship data between the entity data based on the relationship table and constructing a knowledge graph of the procurement object; Based on the existing relationship model of the knowledge graph, a rule queue is set up, and the rule queue is combined with the AMIE+ algorithm to perform rule reasoning and expansion to generate association rules; Expand relational data based on association rules and update the knowledge graph.

7. The enterprise collective purchasing plan recommendation optimization method according to claim 1, characterized in that: The user behavior matrix is used to calculate similarity with multiple association behavior matrices and a valid association matrix is selected. Based on the valid association matrix, the associated purchase objects are marked. The purchase plan is set and recommended in the e-commerce purchase terminal to generate purchase recommendation information. Specifically, The Jordan standard similarity determination method is introduced to calculate the similarity between the user behavior matrix and multiple related behavior matrices; Filter out the correlation behavior matrices whose similarity is greater than a preset threshold to obtain a valid correlation matrix; Perform data analysis on the effective association matrix and mark the associated procurement objects; Based on the associated procurement objects, the e-commerce procurement terminal conducts procurement object attribute analysis and supplier supply status correlation analysis, performs procurement plan setting and recommendation at the e-commerce procurement terminal, and generates procurement recommendation information.

8. An enterprise collective procurement plan recommendation and optimization system, characterized by: The system includes: a memory and a processor, wherein the memory includes an enterprise collective procurement plan recommendation optimization program, and when the enterprise collective procurement plan recommendation optimization program is executed by the processor, the steps of the enterprise collective procurement plan recommendation optimization method as described in any one of claim 1 are implemented.

9. An enterprise collective procurement plan recommendation and optimization platform, characterized by: The platform includes: a data acquisition module, a data processing module, and a visualization module. When the platform is running, the steps of the enterprise collective procurement plan recommendation optimization method according to any one of claims 1 to 7 are implemented.

Images