Intelligent supplier recommendation method and system for shipping industry

Abstract

The invention relates to an intelligent supplier recommendation method and system for the shipping industry. The method comprises the following steps: collecting historical business receipts and real-time business receipts by adopting a data acquisition technology; cleaning the collected business documents; extracting feature data from the cleaned data, converting the feature data into model understandable feature vectors, and respectively constructing an ordering frequency model and an inquiry frequency model; and calculating recommendation scores of output results of the ordering frequency model and the inquiry frequency model based on business weights, and comprehensively recommending a plurality of suppliers meeting requirements in combination with a Top K recommendation algorithm. The system comprises a data acquisition and storage module, a data processing module, a model construction training module and a supplier recommendation module. According to the method and the system provided by the invention, suppliers can be accurately recommended according to industry characteristics and user requirements, data support and reference are provided, the user is helped to make a more intelligent decision, the business processing efficiency is improved, and the risk is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of digital transformation and intelligent development of the shipping industry, and in particular to a supplier intelligent recommendation method and system for the shipping industry. Background Art

[0002] The shipping industry is facing the need for digital transformation and intelligent development. Not only shipping companies, but also centralized procurement companies can improve their overall competitiveness by optimizing supply chain management and improving procurement and supply service levels. Price comparison procurement is a common procurement model for B-side companies. How to select high-quality and low-priced suppliers from a large number of suppliers and quickly obtain supplier quotations determines the probability of winning the bid in the comparison results, which directly affects the operation of the company. In the traditional model, the selection of suppliers depends on human judgment, and there are mainly the following problems:

[0003] (1) There are many types of equipment and spare parts, and the supplier database has a large amount of data, so manual selection is not efficient;

[0004] (2) Human selection lacks objectivity and is prone to favoring certain specific suppliers or ignoring potential high-quality suppliers;

[0005] (3) Human selection is not very accurate and it is not certain to find suppliers with reasonable prices and reliable quality, thus failing to reduce procurement costs.

[0006] In order to solve the above business pain points, some systems have introduced universal supplier recommendation solutions based on historical transaction data, which have solved the problems of low efficiency and objectivity to a certain extent. However, due to different industries and business characteristics, the recommended data is inaccurate, and there are still some problems and challenges in practical applications. In addition, the data collection process relies on historical data. On the one hand, if the quality of historical data is not high (such as incomplete data, single data source, small amount of data, etc.), it will lead to unsatisfactory training results and affect the recommendation results; on the other hand, historical data also has certain biases, such as bias towards certain specific suppliers or ignoring certain potential high-quality suppliers, which affects the objectivity and accuracy of the recommendation results.

[0007] In summary, historical transaction data and recommendation rules directly determine the accuracy and objectivity of the supplier intelligent recommendation results. Data quality issues are mainly divided into two aspects: data source issues and data quality issues: data source issues include a single source that is not diversified enough, data updates are not timely and cannot reflect the latest supplier information and market changes, etc.; data quality issues include data duplication, incompleteness or errors, inconsistent data formats and standards, and are not conducive to data analysis and comparison; historical data may be leaked, tampered with or unauthorized access security issues, resulting in data quality damage and system threats. The problems with recommendation rules mainly include: they are not applicable to the characteristics of the shipping industry, they are overly dependent on a certain data source or feature, the processing of historical data of different categories is not balanced, the recommendation results are not monitored in place, and data bias cannot be discovered and corrected in a timely manner. Summary of the invention

[0008] The present invention solves the problems of poor quality of data related to suppliers in the existing shipping industry and excessive reliance on a certain data source in recommendation rules, resulting in poor accuracy and objectivity of supplier recommendation results. An intelligent supplier recommendation method for the shipping industry is proposed to collect and analyze historical and real-time business documents, intelligently recommend suitable suppliers, improve procurement and supply efficiency, and avoid the risk of manual intervention.

[0009] The technical solution claimed in the present invention is as follows:

[0010] A supplier intelligent recommendation method for the shipping industry includes the following steps:

[0011] S1: Using data collection technology to collect historical business documents and real-time business documents generated from multiple business systems, and uniformly collect and store the business documents; the business documents include inquiry forms and purchase orders;

[0012] S2: Clean the business documents collected in S1 to remove duplicate data, missing values ​​and outliers;

[0013] S3: extract feature data from the data processed by S2, and convert the feature data into feature vectors that can be understood by the model, respectively construct an order frequency model and an inquiry frequency model, and respectively use the feature vectors corresponding to the order form and the inquiry form to train and optimize the order frequency model and the inquiry frequency model;

[0014] S4: Based on the business weight, the row recommendation score of the output results of the order frequency model and the inquiry frequency model described in S3 is calculated, and combined with the Top K recommendation algorithm, multiple suppliers that meet the needs are comprehensively recommended.

[0015] Preferably, the data collection technology described in S1 includes API interface, real-time data collection, and distributed data collection.

[0016] Preferably, the cleaning in S2 is performed based on the unique identifier and key data fields of the business document to remove duplicate data; and missing values ​​and abnormal values ​​in the business document are detected by a linear traversal method in S2.

[0017] Preferably, the supplier intelligent recommendation method also includes: visually displaying the suppliers obtained in S4, using a front-end page to display the supplier recommendation results, and supporting returning relevant attribute data corresponding to the recommended suppliers according to different page display requirements.

[0018] In the above method, the characteristic data corresponding to the inquiry form in S3 include: ship, equipment, spare parts, supplier, time, inquiry, and the inquiry frequency model inputs the characteristic vector corresponding to the inquiry form and outputs the supplier recommendation result; the characteristic data corresponding to the purchase order include: ship, equipment, spare parts, supplier, time, order; the order frequency model inputs the characteristic vector corresponding to the purchase order and outputs the supplier recommendation result.

[0019] In the above method, the training and optimization described in S3 refers to dividing the data obtained in S2 into a training set, a validation set and a test set, selecting a suitable loss function and optimization algorithm, adjusting model parameters, and performing model evaluation and performance optimization.

[0020] In the above method, S4 is specifically: for the supplier recommendation results output by the ordering frequency model and the inquiry frequency model described in S3, the business weights are set according to the requirements of different business scenarios, and the recommendation score of each supplier in the supplier recommendation results is calculated according to the business weights. Based on the recommendation scores, a Top K recommendation algorithm is used to screen multiple suppliers that meet the requirements for recommendation.

[0021] The specific calculation formula of the recommendation score in the above method is:

[0022] Recommendation score = option frequency × time decay factor;

[0023] The calculation formula of time decay factor TDF(t) is:

[0024]

[0025] Where: λ is the decay rate; t is the current time; t0 is the time when the event occurs;

[0026] The formula for calculating the frequency of a choice is:

[0027] FBRSS=ISFi×TDF(t)

[0028] Wherein: ISFi is the frequency of the participating items; the frequency of the participating items includes: ordering frequency and inquiry frequency.

[0029] Preferably, the order frequency and inquiry frequency are obtained by respectively counting the number of orders and the number of inquiries within a certain time period.

[0030] The present invention provides an intelligent supplier recommendation system for the shipping industry, comprising a data collection and storage module connected in sequence and using data collection technology to collect historical business documents and real-time business documents generated from multiple business systems and to uniformly aggregate and store the business documents, a data processing module used to clean the business documents collected by the data collection and storage module and remove duplicate data, missing values ​​and abnormal values, a model construction training module used to extract feature data from data processed by the data processing module and convert the feature data into feature vectors understandable to the model and respectively construct an order frequency model and an inquiry frequency model, and a supplier recommendation module used to calculate the recommendation scores of the output results of the order frequency model and the inquiry frequency model in the training module based on business weights and comprehensively recommend multiple suppliers that meet the needs in combination with a Top K recommendation algorithm; the business documents include an inquiry form and a purchase form; the order frequency model and the inquiry frequency model respectively use the feature vectors corresponding to the purchase form and the inquiry form to train the model.

[0031] Beneficial Effects

[0032] The present invention provides a method and system for intelligently recommending suppliers for the shipping industry. The method comprises: using data acquisition technology to collect historical business documents and real-time business documents generated from multiple business systems, ensuring the timeliness and accuracy of data, providing a basis for subsequent analysis and recommendation, and uniformly collecting and storing business documents; cleaning the collected business documents, removing duplicate data, eliminating missing values ​​and abnormal values, ensuring the integrity and accuracy of data, improving data quality, solving the problem of poor data quality related to suppliers in the existing shipping industry, and improving the accuracy of subsequent model training; processing and converting the data obtained by the above processing according to industry characteristics and data characteristics, extracting feature features meaningful for modeling and prediction, and converting the feature data into feature vectors that can be understood by the model, respectively constructing an order frequency model and an inquiry frequency model, and respectively using the feature vectors corresponding to the order form and the inquiry form to train the order frequency model and the inquiry frequency model; calculating the recommendation scores of the output results of the order frequency model and the inquiry frequency model based on the business weight, and combining the Top The K recommendation algorithm comprehensively recommends multiple suppliers that meet the needs, uses inquiry forms and purchase orders to train the models respectively, and comprehensively considers the recommendation results of the two models to select suppliers, thereby solving the problem that the existing supplier recommendation rules in the shipping industry are overly dependent on a certain data source. Through the above method, the problems of poor data quality related to existing shipping industry suppliers and excessive dependence of recommendation rules on a certain data source, resulting in poor accuracy and objectivity of supplier recommendation results, are solved.

[0033] The data collection technology includes an API interface. Collecting data through the API interface can ensure the real-time and accuracy of the data.

[0034] The cleaning is performed based on the unique identifier and key data fields of the business document to remove duplicate and redundant data, which can further improve the quality of the business document and provide accuracy for subsequent models.

[0035] The supplier intelligent recommendation method also includes: visually displaying the suppliers obtained in S4, using a front-end page to display the supplier recommendation results, and supporting returning relevant attribute data corresponding to the recommended suppliers according to different page display requirements, so as to meet the diversified business needs and improve the user experience.

[0036] The method and system provided by the present invention build a supplier intelligent recommendation model and application with the characteristics of the shipping industry, which can accurately recommend suppliers according to industry characteristics and user needs, provide data support and reference, help users make more informed decisions, improve business processing efficiency, and reduce risks. At the same time, the maturity of intelligent recommendation technology can be reused in other supply chain businesses in the shipping industry, empowering the digitalization of the shipping industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a flow chart of a supplier intelligent recommendation method for the shipping industry according to an embodiment of the present invention.

[0038] Figure 2 Schematic diagram of an intelligent supplier recommendation system for the shipping industry according to an embodiment of the present invention.

[0039] Figure 3 Front-end display page for supplier recommendation results. Specific implementation methods

[0041] In order to make the purpose, technical solution and advantages of the present invention more clear, the technical solution is further clearly and completely described below in conjunction with the accompanying drawings of the present invention.

[0042] The following embodiment is a supplier intelligent recommendation model and application comprehensive solution based on data collection and data modeling. First, a data collection mechanism is established according to business characteristics to collect raw data as the accumulation of recommendation data. Second, the collected raw data is converted into a form that can be used for analysis and recommendation. Finally, a recommendation system service is built to support integration with different business systems. The overall architecture of the adopted system technology stack is as follows:

[0043] Based on the Spring Cloud microservice architecture, the entire system is divided into multiple small independent services, each with specific functions and responsibilities; PostgreSQL is used as the database for data storage and management; Jenkins is used to implement continuous integration and continuous deployment (CI / CD) to simplify the development process and improve delivery efficiency; Kubernetes (K8s) is used as a container orchestration and management tool to achieve high availability, automatic expansion and deployment management; Nginx is used as a reverse proxy server to achieve load balancing and secure access; Ant Design Vue (Antdv) is used as the front-end to provide users with a friendly interface and data visualization functions; Vue 3 is used as the main front-end development framework to provide a responsive user interface and powerful componentization capabilities; Ant Design Vue (Antdv) is used as a UI component library to provide rich UI components and styles; Node.js is used as the back-end JavaScript runtime environment to provide support for front-end building, packaging and deployment.

[0044] The inventor of this application investigated the characteristics and needs of the procurement and supply business scenarios in the shipping industry, and gained an in-depth understanding of the basis for selecting suppliers when purchasing / inquiring for various products in different business lines. He communicated and interviewed customers such as shipowners and operators to collect their needs and pain points. He analyzed the needs and problems, clarified the goals and directions of the research, and provided a supplier intelligent recommendation method and system for the shipping industry.

[0045] The first set of embodiments: intelligent supplier recommendation method for the shipping industry

[0046] This group of embodiments provides a supplier intelligent recommendation method for the shipping industry, such as Figure 1 As shown, the following steps are included:

[0047] S1: Use data collection technology to collect historical business documents and real-time business documents generated from multiple business systems, and uniformly collect and store the business documents; the business documents include inquiry forms and purchase orders; the data collection technology includes API interface, real-time data collection, and distributed data collection;

[0048] S2: Clean the business documents collected in S1 to remove duplicate data, missing values ​​and abnormal values; in a specific embodiment of the present invention, the cleaning is performed based on the unique mark and key data field of the business document to remove duplicate data. Each business document corresponds to a unique identification code (unique mark). If the unique identification codes of two business documents are the same, then one of the business documents is removed. Similarly, the value of the key data field in each business document is unique. If the value of the key data field of two business documents is the same, then one of the business documents is removed; missing values ​​and abnormal values ​​in the business documents are detected by a linear traversal method, and the missing values ​​and abnormal values ​​are removed.

[0049] S3: Extract feature data from the data processed by S2, and convert the feature data into feature vectors that can be understood by the model, respectively construct an ordering frequency model and an inquiry frequency model, and respectively use the feature vectors corresponding to the order form and the inquiry form to train and optimize the ordering frequency model and the inquiry frequency model; the feature data corresponding to the inquiry form include: ship, equipment, spare parts, supplier, time, inquiry, the inquiry frequency model inputs the feature vector corresponding to the inquiry form, and outputs the supplier recommendation result; the feature data corresponding to the order form include: ship, equipment, spare parts, supplier, time, order; the ordering frequency model inputs the feature vector corresponding to the order form, and outputs the supplier recommendation result; the training and optimization refers to dividing the data obtained by S2 into a training set, a validation set and a test set, selecting a suitable loss function and optimization algorithm, adjusting the model parameters, and performing model evaluation and performance optimization.

[0050] S4: Based on the business weight, calculate the recommendation score of the supplier recommendation results output by the ordering frequency model and the inquiry frequency model described in S3, and combine with the Top K recommendation algorithm to comprehensively recommend multiple suppliers that meet the needs; specifically: obtain multiple preliminary supplier recommendation results (multiple suppliers) based on the output results of the ordering frequency model and the inquiry frequency model described in S3, set business weights according to the requirements of different business scenarios, and calculate the recommendation score of each supplier in the supplier recommendation results according to the business weight, and based on the recommendation score, use the Top K recommendation algorithm to screen multiple suppliers that meet the needs for recommendation.

[0051] The specific calculation formula of the recommendation score is:

[0052] Recommendation score = option frequency × time decay factor;

[0053] Where: The calculation formula of time decay factor TDF(t) is:

[0054]

[0055] Where: λ is the decay rate; t is the current time; t0 is the time when the event occurs; the calculation formula for the option frequency is:

[0056] FBRSS=ISFi×TDF(t)

[0057] Wherein: ISFi is the frequency of the participating items; the frequency of the participating items includes: order frequency and inquiry frequency; the order frequency and inquiry frequency are obtained by counting the number of orders and the number of inquiries in a certain time period respectively.

[0058] In a specific embodiment of the present invention, when calculating the recommendation score, the weight is reflected in the number of orders from different suppliers or the number of inquiries from different suppliers. Assuming that a certain supplier focuses on the ordering frequency, then when calculating the recommendation score of this supplier, the weight of the ordering frequency is greater than the inquiry frequency; on the contrary, assuming that a certain supplier focuses on the inquiry frequency, then when calculating the recommendation score of this supplier, the weight of the inquiry frequency is greater than the ordering frequency.

[0059] In a specific embodiment of the present invention, multiple recommended suppliers can be obtained according to the output results of the ordering frequency model and the inquiry frequency model described in S3, each supplier corresponds to a recommendation score, and a Top K recommendation algorithm is used to sort the suppliers from large to small according to the recommendation scores, and multiple suppliers ranked in front (with high recommendation scores) are selected as the most suitable suppliers for recommendation.

[0060] S5: Visualize the suppliers obtained in S4, use the front-end page to display the supplier recommendation results, and support returning the relevant attribute data corresponding to the recommended suppliers according to different page display requirements; Figure 3 It is the top N recommended data (suppliers) obtained after S4 processing.

[0061] Second set of embodiments: Intelligent supplier recommendation system for the shipping industry

[0062] This group of embodiments provides a supplier intelligent recommendation system for the shipping industry. Figure 2As shown, it includes a data acquisition storage module connected in sequence and using data acquisition technology to collect historical business documents and real-time business documents generated from multiple business systems and uniformly collect and store the business documents, a data processing module for cleaning the business documents collected by the data acquisition storage module and removing duplicate data, missing values ​​and abnormal values, a model construction training module for extracting feature data from the data processed by the data processing module and converting the feature data into feature vectors that can be understood by the model and respectively constructing an order frequency model and an inquiry frequency model, and a supplier recommendation module for calculating the recommendation scores of the output results of the order frequency model and the inquiry frequency model in the training module based on the business weight calculation model and combining the Top K recommendation algorithm to comprehensively recommend multiple suppliers that meet the needs; the business documents include inquiry forms and purchase orders; the order frequency model and the inquiry frequency model respectively use the feature vectors corresponding to the purchase order and the inquiry form to train the model.

[0063] Use historical data to verify and optimize the models established by the above methods and systems to ensure the accuracy and reliability of the analysis results; further optimize and improve the models in the above methods and systems based on user feedback and opinions. Perform performance testing and security audits to ensure the stability and availability of the technical solutions.

[0064] Provide training and technical support to procurement business operator users to ensure that they can use the above methods and systems smoothly. For different procurement and supply chain businesses, first conduct pilot applications on spare parts procurement and supply business, and then promote them to other procurement and supply business scenarios, and continuously iterate and optimize products in practice. Promote application in the shipping industry, expand influence, collect user feedback, and continuously improve and upgrade products.

[0065] The above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the present invention. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the present invention.

Claims

1. A supplier intelligent recommendation method for the shipping industry, characterized in that: The steps include: S1: Using data collection technology to collect historical business documents and real-time business documents generated from multiple business systems, and uniformly collect and store the business documents; the business documents include inquiry forms and purchase orders; S2: Clean the business documents collected in S1 to remove duplicate data, missing values ​​and outliers; S3: extract feature data from the data processed by S2, and convert the feature data into feature vectors that can be understood by the model, respectively construct an order frequency model and an inquiry frequency model, and respectively use the feature vectors corresponding to the order form and the inquiry form to train and optimize the order frequency model and the inquiry frequency model; S4: Based on the business weight, the recommendation score of the output results of the order frequency model and the inquiry frequency model described in S3 is calculated, and combined with the Top K recommendation algorithm, multiple suppliers that meet the needs are comprehensively recommended.

2. The intelligent supplier recommendation method for the shipping industry according to claim 1, characterized in that: The data collection technology described in S1 includes API interface, real-time data collection, and distributed data collection.

3. The intelligent supplier recommendation method for the shipping industry according to claim 1, characterized in that: The cleaning described in S2 is performed based on the unique identifier and key data fields of the business document to remove duplicate data; in S2, missing values ​​and abnormal values ​​in the business document are detected by a linear traversal method.

4. The intelligent supplier recommendation method for the shipping industry according to claim 1, characterized in that: The supplier intelligent recommendation method also includes: visually displaying the suppliers obtained in S4, using a front-end page to display the supplier recommendation results, and supporting returning relevant attribute data corresponding to the recommended suppliers according to different page display requirements.

5. The intelligent supplier recommendation method for the shipping industry according to claim 1, characterized in that: The characteristic data corresponding to the inquiry form in S3 include: ship, equipment, spare parts, supplier, time, inquiry, and the inquiry frequency model inputs the characteristic vector corresponding to the inquiry form and outputs the supplier recommendation result; the characteristic data corresponding to the purchase order include: ship, equipment, spare parts, supplier, time, order; the order frequency model inputs the characteristic vector corresponding to the purchase order and outputs the supplier recommendation result.

6. The intelligent supplier recommendation method for the shipping industry according to claim 1, characterized in that: The training and optimization described in S3 refers to dividing the data obtained in S2 into a training set, a validation set, and a test set, selecting a suitable loss function and optimization algorithm, adjusting model parameters, and performing model evaluation and performance optimization.

7. The intelligent supplier recommendation method for the shipping industry according to claim 5, characterized in that: S4 is specifically as follows: for the supplier recommendation results output by the ordering frequency model and the inquiry frequency model described in S3, business weights are set according to the demands of different business scenarios, and the recommendation score of each supplier in the supplier recommendation results is calculated according to the business weights. Based on the recommendation scores, a Top K recommendation algorithm is used to screen multiple suppliers that meet the requirements for recommendation.

8. The intelligent supplier recommendation method for the shipping industry according to claim 7, characterized in that: The specific calculation formula of the recommendation score is: Recommendation score = option frequency × time decay factor; The calculation formula of time decay factor TDF(t) is: Where: λ is the decay rate; t is the current time; t0 is the time when the event occurs; The formula for calculating the frequency of a choice is: RBRS ui =ISF i ×TDF(t) Wherein: ISFi is the frequency of the participating items; the frequency of the participating items includes: ordering frequency and inquiry frequency.

9. The intelligent supplier recommendation method for the shipping industry according to claim 8, characterized in that: The order frequency and inquiry frequency are obtained by respectively counting the number of orders and the number of inquiries within a certain period of time.

10. An intelligent supplier recommendation system for the shipping industry, characterized in that: The invention comprises a data collection and storage module which is connected in sequence and uses data collection technology to collect historical business documents and real-time business documents generated from multiple business systems and uniformly aggregate and store the business documents, a data processing module which is used to clean the business documents collected by the data collection and storage module and remove duplicate data, missing values ​​and abnormal values, a model construction and training module which is used to extract feature data from the data processed by the data processing module and convert the feature data into feature vectors understandable to the model and respectively construct an order frequency model and an inquiry frequency model, and a supplier recommendation module which is used to calculate the recommendation scores of the output results of the order frequency model and the inquiry frequency model in the training module based on the business weight calculation model and comprehensively recommend multiple suppliers that meet the demand in combination with the Top K recommendation algorithm; the business documents include inquiry forms and purchase forms; the order frequency model and the inquiry frequency model respectively use the feature vectors corresponding to the purchase form and the inquiry form to train the model.

Images