Sandstone information personalized recommendation method and platform based on user transaction behavior

Abstract

The application discloses a sandstone information personalized recommendation method and platform based on user transaction behavior, and relates to the technical field of information recommendation.The application solves the technical problem that a systematic recommendation effect quantification system is lacking and active recommendation is difficult to achieve.The application builds a user portrait containing time, preference and behavior sequence characteristics, filters similar users based on multi-dimensional characteristics, improves the matching accuracy of a recommendation scheme and user demand, introduces user behavior sequence characteristic analysis, combines algorithms such as LSTM to capture demand change trends, and formulates differentiated recommendation strategies for fluctuating purchase quantities.Compared with a static recommendation mode, the method can improve the repeat purchase rate of purchasers, effectively cope with industry periodic demand fluctuations, establish a multi-dimensional quantification evaluation system for accuracy, conversion rate and user satisfaction, and sort recommendation schemes through weighted calculation of comprehensive effect indicators.The system can improve the iteration efficiency of the recommendation system and continuously optimize the user experience compared with single indicator evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information recommendation, in particular to a sand and gravel information personalized recommendation method and platform based on user transaction behavior. BACKGROUND

[0002] In the digital transaction scenario of the sand and gravel industry, traditional information recommendation methods mostly rely on general rules or simple user-item matching logic, such as keyword-based search or popular product recommendation.

[0003] According to the patent application with publication number CN118568335A, a real-time personalized recommendation system is disclosed, which includes a data collection module, a real-time data processing module, a user modeling module, a recommendation algorithm module, a feedback processing module, a recommendation result display module, and a real-time updating module. The present application provides a real-time personalized recommendation system, which can better model user interest in goods / services by using the TF-IDF algorithm.

[0004] However, traditional recommendation methods do not build a multi-dimensional feature model of user transaction behavior, cannot identify user personalized needs, often have a mismatch between recommended solutions and actual needs, making it difficult for buyers to quickly obtain suitable sand and gravel resources. The procurement of sand and gravel is greatly affected by policy and market price fluctuations. The existing technology cannot effectively use historical transaction sequences to predict user demand changes, making it difficult to achieve proactive recommendation. There is a lack of systematic recommendation effect quantification system, and the effectiveness of the scheme cannot be evaluated from multiple dimensions such as accuracy, conversion rate, and user satisfaction. It is not possible to continuously optimize the recommendation strategy based on feedback, resulting in poor user experience and waste of supplier resources. SUMMARY

[0005] To overcome the shortcomings of the prior art, the present application provides a sand and gravel information personalized recommendation method and platform based on user transaction behavior, which solves the problem of lack of systematic recommendation effect quantification system and difficulty in achieving proactive recommendation.

[0006] To achieve the above purpose, the present application is implemented by the following technical scheme: a sand and gravel information personalized recommendation method based on user transaction behavior, which specifically includes the following steps:

[0007] Collect user transaction information and perform data cleaning and desensitization to obtain preprocessed data, establish a user portrait, and extract time features, preference features, and behavior sequence features;

[0008] Filter similar users and corresponding procurement solutions based on the user portrait, and filter the procurement solutions based on the extracted features to obtain preselected solutions;

[0009] The preselected scheme is twice screened based on the sandstone type corresponding to the behavior sequence feature in the user portrait, an initial selected scheme is obtained, and the purchase scheme is screened based on the non-purchase record to obtain a scheme to be analyzed;

[0010] The associated article scheme is obtained by screening the initial selected scheme based on the associated article corresponding to the user's purchased record and taking the associated article as a standard, and the effect index is obtained by comprehensively calculating the accuracy, conversion rate and user satisfaction index of the associated article scheme;

[0011] According to the effect index, the purchase change signal or the purchase change signal is generated by analyzing the user cycle purchase situation, the maximum effect index scheme meeting the user area is screened to generate the recommendation information for the purchase change signal;

[0012] For the purchase change signal, the specific change condition is determined, the increase or decrease analysis signal is generated, and the recommendation information is generated by analyzing respectively.

[0013] As a further scheme of the present application, the specific way of screening the purchase scheme to obtain the preselected scheme is:

[0014] The user portrait and the corresponding features are obtained, and the user with an intersection feature is recorded as a similar user, and the corresponding purchase scheme is recorded as i, and i=1, 2, …, j, wherein j represents the number of purchase schemes, and the sandstone label of the purchase scheme i is established, the purchase scheme i is screened based on the features corresponding to the user portrait, the preselected scheme is obtained and labeled as a, and a=1, 2, …, b, wherein b represents the number of preselected schemes.

[0015] As a further scheme of the present application, the specific way of obtaining the scheme to be analyzed is:

[0016] The behavior sequence feature corresponding to the user portrait is obtained, and the purchased record and the non-purchased record corresponding to the behavior sequence feature are obtained, and the preselected scheme a is twice screened according to the sandstone type corresponding to the purchased record to obtain the initial selected scheme, then the sandstone type corresponding to the non-purchased record is determined and recorded as the analyzed type, and all purchase schemes are re-screened according to the analyzed type to obtain the scheme to be analyzed, recorded as n, and n=1, 2, …, m, wherein m represents the number of schemes to be analyzed.

[0017] As a further scheme of the present application, the specific way of comprehensively calculating the effect index is:

[0018] The initial selected scheme and the user's purchased record are obtained, the associated article is determined according to the easy-to-buy record, and the associated article represents the article that exists in combination purchase, the associated article scheme is obtained by screening the initial selected scheme based on the associated article, and the corresponding quantitative index is obtained by analyzing from the accuracy, conversion rate and user satisfaction.

[0019] According to the formula effect index = accuracy index * first weight + conversion rate index * second weight + satisfaction index * third weight, the effect index of the associated item scheme is calculated.

[0020] As a further scheme of the application, the way of obtaining the corresponding quantitative index from the three aspects of accuracy, conversion rate and user satisfaction is:

[0021] The number of associated item schemes matching the user portrait is obtained, according to the formula matching proportion = matching number ÷ total number * 100%, and according to the formula repeat purchase ratio = repeat purchase times ÷ total recommendation times * 100%, then according to the formula accuracy = matching proportion * z1 + repeat purchase ratio * z2, the accuracy index is calculated, and z1 and z2 are corresponding weight ratio values;

[0022] The number of times the user clicks the associated item scheme is obtained, and according to the formula click conversion rate = click times ÷ total recommendation times * 100%, and according to the formula purchase conversion rate = purchase times after recommendation ÷ total recommendation times * 100%, and according to the formula conversion rate = click conversion rate * q1 + purchase conversion rate * q2, the conversion rate index is calculated, and q1 and q2 are corresponding weight ratio values;

[0023] The average score obtained directly through user research, and according to the formula satisfaction score = total score ÷ number of scores, and the satisfaction score obtained is standardized to obtain the score index, and all the number of good feedback is obtained, and according to the formula feedback good index = (number of good feedback ÷ total number of feedback), then the formula satisfaction index = score index * r1 + feedback good index * r2, the satisfaction index is calculated, and r1 and r2 are corresponding weight ratio values.

[0024] As a further scheme of the application, the specific way of generating recommendation information by screening the maximum effect index scheme that meets the user area is:

[0025] According to the effect index from large to small, the associated item schemes are sorted, and the current user cycle procurement situation is obtained, if the cycle procurement situation is unchanged, a procurement unchanged signal is generated, otherwise if the procurement situation is changed, a procurement change analysis signal is generated;

[0026] The procurement unchanged signal is analyzed, and the associated item scheme corresponding to the maximum effect index that meets the region is screened according to the user's region, and the recommendation information is generated according to the same.

[0027] As a further scheme of the application, the analysis way for the procurement change signal is:

[0028] The acquisition period procurement specific situation is generated to generate an increase analysis signal or a decrease analysis signal, the obtained increase analysis signal is processed, the first two positions of the effect index are selected as a standard to screen the associated item scheme, and the recommendation information is generated. The decrease analysis signal is processed, all analysis schemes n are obtained, the most frequently viewed sandstone type corresponding to the behavior sequence feature in the user portrait is obtained, and the most frequently viewed sandstone type is used as a standard to screen the analysis scheme n, and the user area and the purchased record are combined for comprehensive analysis to screen the analysis scheme to generate the recommendation information.

[0029] The sandstone information personalized recommendation platform based on user transaction behavior comprises:

[0030] The user information acquisition module is used for collecting user transaction information and preprocessing to obtain preprocessed data, and transmitting the preprocessed data to the user portrait establishment module.

[0031] The user portrait establishment module is used for establishing a user portrait according to the obtained preprocessed data, extracting corresponding features of the user, including time features, preference features and sequence features, screening similar users according to the user portrait, screening preselected schemes from the purchase schemes of the similar users, and then transmitting the preselected schemes to the scheme processing and analysis module.

[0032] The scheme processing and analysis module is used for screening the preselected schemes according to the behavior sequence features in the user portrait to obtain preliminary selected schemes, screening all purchase schemes according to the sandstone types to obtain analysis schemes, determining associated items according to the user behavior sequence features, and screening the associated item schemes from the preliminary selected schemes according to the associated items as a standard. Then, the effect index is comprehensively calculated from the accuracy, the conversion rate and the user satisfaction, and is transmitted to the scheme processing and analysis module.

[0033] The scheme processing and analysis module is used for sorting according to the effect index, analyzing the user periodic purchase situation to generate a purchase invariable signal or a purchase change signal, screening the maximum effect index scheme meeting the user area to generate the recommendation information for the purchase invariable signal, determining the specific change situation to generate an increase or decrease analysis signal for the purchase change signal.

[0034] The increase analysis signal is processed, and the associated item scheme is screened according to the maximum two effect indexes to generate the recommendation information.

[0035] The decrease analysis signal is processed, the most frequently viewed sandstone type corresponding to the behavior sequence feature is obtained, and the analysis scheme is screened in combination with the user area and the purchased record to generate the recommendation information, and the generated recommendation information is transmitted to the recommendation information display module.

[0036] The recommendation information display module is used for displaying the obtained recommendation information to the corresponding user.

[0037] The present application provides a sand and gravel information personalized recommendation method and platform based on user transaction behavior. Compared with the prior art, the following beneficial effects are achieved:

[0038] The present application preprocesses user transaction information, constructs a user portrait containing time, preference and behavior sequence features, and generates a recommendation scheme based on multi-dimensional feature screening of similar users and associated item analysis. Compared with traditional methods, this technology can improve the matching accuracy of the recommendation scheme and user demand, and reduce the time and cost of screening by the purchaser.

[0039] The present application introduces user behavior sequence feature analysis, captures demand change trends using algorithms such as LSTM, and develops differentiated recommendation strategies for fluctuating procurement quantities. Compared with static recommendation mode, this method can improve the repeat purchase rate of purchasers to effectively respond to cyclical demand fluctuations in the industry, establish a multi-dimensional quantitative evaluation system for accuracy, conversion rate and user satisfaction, and sort the recommendation scheme by weighted calculation of comprehensive effect indicators. Compared with single index evaluation, this system can improve the iteration efficiency of the recommendation system and continuously optimize the user experience. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 The present application is a step-by-step method diagram;

[0041] Figure 2 The present application is a recommendation platform block diagram. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0043] Embodiment one

[0044] Please refer to Figure 1 The present application provides a sand and gravel information personalized recommendation method based on user transaction behavior, which specifically includes the following steps:

[0045] Step S1, obtain user transaction information and perform preprocessing operation to obtain preprocessed data, and the preprocessing operation includes data cleaning and data desensitization;

[0046] Specific data cleaning: remove abnormal transaction data (such as test orders, repeated records), and fill in missing values (such as filling in prices by interpolating adjacent transaction data);

[0047] Data desensitization: encrypt storage of user privacy information (such as enterprise contact information);

[0048] And according to the obtained preprocessed data, the corresponding user portrait is established, and the corresponding time characteristics, preference characteristics and behavior sequence characteristics are extracted according to the user portrait.

[0049] Specific time characteristics: (purchase time interval, seasonal purchase rule, etc.);

[0050] Preference characteristics: (commonly used sandstone types, specifications, price sensitivity, etc.).

[0051] Behavior sequence characteristics: (historical transaction sequence, browsing track, etc.).

[0052] Step S2, based on the user portrait, from the time characteristics (purchase time interval, seasonal purchase rule, etc.), preference characteristics (commonly used sandstone types, specifications, price sensitivity, etc.), behavior sequence characteristics (historical transaction sequence, browsing track, etc.) Three dimensions of screening similar users;

[0053] Determination rule: as long as the user has any intersection in the time characteristics, preference characteristics and behavior sequence characteristics, it is determined as a similar user. For example, user A and user B both commonly use "specification-medium sand" in sandstone preference characteristics, even if there is no intersection in time characteristics and behavior sequence characteristics, user A and user B are also recognized as similar users.

[0054] Get all similar user corresponding purchase scheme, and mark the order number as i (i=1, 2, …, j, j is the total number of purchase scheme), for each purchase scheme i, according to the sandstone attribute, establish the corresponding sandstone label, such as "origin-Hubei", "specification-medium sand", "use-concrete", take the time characteristics, preference characteristics and behavior sequence characteristics corresponding to the current user portrait as the screening standard, screen the purchase scheme i. Only the purchase scheme that matches the three characteristics of the current user can be selected as the preselected scheme, and the number is marked as a (a=1, 2, …, b, b is the total number of preselected scheme);

[0055] Extract the behavior sequence characteristics in the user portrait, and clarify the purchased records and the non-purchased records (the non-purchased records include only browsing but not purchasing sandstone information), and perform secondary screening on the preselected scheme a according to the sandstone type corresponding to the purchased record, and retain the scheme matching the sandstone type to obtain the initial selected scheme. Organize the sandstone types corresponding to the non-purchased records, and mark them as types to be analyzed. According to the types to be analyzed, all purchase schemes (including schemes not entering the preselected stage) are re-screened, and the schemes meeting the types to be analyzed are determined as the schemes to be analyzed, and the number is marked as n (n=1, 2, …, m, m is the total number of schemes to be analyzed).

[0056] Suppose there is a user X, whose user portrait is as follows:

[0057] Time feature: The procurement cycle is once a month, and the procurement peak is concentrated in March to June.

[0058] Preference feature: The commonly used sand type is "river sand - medium sand", the price sensitivity is medium, and it is mainly used for concrete mixing.

[0059] Behavior sequence feature: In the purchase record in the past half year, 80% of the river sand is "origin - Jiangxi, specification - medium sand, use - concrete"; in the non-purchase record, the user has browsed the river sand "origin - Hubei, specification - medium sand, use - concrete" many times, but no purchase behavior has been generated.

[0060] The system finds that the commonly used sand type of user Y is also "river sand - medium sand", although the time feature and behavior sequence feature are different from user X, but still determines that user Y is a similar user, and obtains the procurement scheme of user Y, in the obtained multiple similar user procurement schemes, the sandstone label is established. According to the three features of user X as the standard for screening, only the procurement scheme that meets the conditions such as "procurement once a month, commonly used river sand - medium sand for concrete mixing" can become a pre-selected scheme, according to the type of "Jiangxi river sand - medium sand" in the purchase record of user X, the pre-selected scheme is screened again to obtain the initial selected scheme; for the type to be analyzed "Hubei river sand - medium sand" in the non-purchase record of user X, all procurement schemes are re-screened to obtain the analysis scheme.

[0061] Step S3, obtaining the initial selected scheme and the analysis scheme, then analyzing the corresponding associated items according to the purchase record of the user, and the associated items here represent the items that exist in combination purchase, and at the same time, the associated items are used as the standard to screen the initial selected scheme to obtain the associated item scheme, and the scheme effect index of the associated item scheme is calculated, and the accuracy, conversion rate and user satisfaction are specifically analyzed;

[0062] The accuracy is analyzed, the number of associated item schemes matched with the user portrait is obtained, and the corresponding matching proportion is calculated according to the formula matching proportion = matching number ÷ total number × 100%, and the repurchase proportion of the sandstone type that has been purchased in the recommended scheme is calculated according to the formula repurchase proportion = repurchase times ÷ total recommendation times × 100%, then the accuracy is calculated according to the formula accuracy = matching proportion × z1 + repurchase proportion × z2, and the accuracy index is obtained, and z1 and z2 are the corresponding weight ratio;

[0063] The conversion rate is analyzed, the number of times that the user clicks the associated item scheme is obtained, and the formula click conversion rate = click number ÷ total recommendation number × 100% is used, the order number ratio of the associated item scheme completing the purchase is obtained, the formula purchase conversion rate = purchase number after recommendation ÷ total recommendation number × 100% is used, and the conversion rate index is calculated according to the formula conversion rate = click conversion rate × q1 + purchase conversion rate × q2, and q1 and q2 are corresponding weight ratio values;

[0064] The user satisfaction is analyzed, the average score directly obtained through user investigation (such as 1-5 score) is obtained, the formula satisfaction score = total score ÷ score number is used, the obtained satisfaction score is standardized to obtain the score index, specifically 4 / 5 obtains 0.8, the number of all good feedbacks is obtained, and the good feedback here means the investigation with a score of 5, the formula feedback good index = (good feedback number ÷ total feedback number) is used, then the score index and the feedback good index are weighted and summed, the formula satisfaction index = score index × r1 + feedback good index × r2 is used, and the satisfaction index is calculated, wherein r1 and r2 are corresponding weight ratio values;

[0065] In step S4, the accuracy index, the conversion rate index and the satisfaction index are weighted and summed, the formula effect index = accuracy index × first weight + conversion rate index × second weight + satisfaction index × third weight is used, the effect index of the associated item scheme is calculated, the associated item schemes are sorted according to the effect index from large to small, the current user cycle purchase situation is obtained, and the cycle purchase situation here means that the purchase amount change of the previous purchase cycle is obtained with time T as the cycle; if the cycle purchase situation is unchanged, a purchase unchanged signal is generated, otherwise if the purchase situation is changed, a purchase change analysis signal is generated;

[0066] For the generated purchase unchanged signal, the associated item scheme corresponding to the maximum effect index of the region meeting the user's region is selected as the standard to generate the recommendation information, specifically, the associated item scheme meeting the user's region is first determined, and then the associated item scheme with the maximum effect index is selected;

[0067] The generated purchase change signal is analyzed, the specific situation of the periodic purchase is obtained, an increase analysis signal or a decrease analysis signal is generated, the obtained increase analysis signal is processed, the first two positions of the effect index are selected as a standard to screen the associated item scheme, and the recommended information is generated. The decrease analysis signal is processed, all to-be-analyzed schemes n are obtained, the corresponding most browsing times of sandstone types in the behavior sequence features in the user portrait are obtained, and the to-be-analyzed schemes n are screened according to the sandstone types as a standard. The user area and the purchased record are comprehensively analyzed, the to-be-analyzed schemes are screened to generate recommended information. Specifically, the to-be-analyzed schemes are first screened according to the sandstone types to obtain preliminary screening to-be-analyzed schemes. Then, the to-be-analyzed schemes are secondarily screened according to the user area to obtain secondary screening to-be-analyzed schemes. Finally, the user purchased record is obtained, and the secondary screening to-be-analyzed schemes are finally screened according to the most purchased times in the purchased record as a standard to determine the corresponding recommended standard.

[0068] Embodiment two

[0069] Please refer to Figure 2 The application provides a sandstone information personalized recommendation platform based on user transaction behavior, which comprises a user information acquisition module, a user portrait establishment module, a scheme processing and analysis module, a comprehensive screening and determination module, and a recommended information output module.

[0070] The user information acquisition module is used for acquiring user transaction information and pre-processing to obtain pre-processed data, and simultaneously transmitting the pre-processed data to the user portrait establishment module.

[0071] The user portrait establishment module is used for establishing a user portrait according to the obtained pre-processed data, extracting corresponding features of the user, including time features, preference features, and sequence features, screening similar users according to the user portrait, screening preselected schemes from the purchase schemes of the similar users, and then transmitting the preselected schemes to the scheme processing and analysis module. The specific processing manner is the same as the processing process of step S2 in embodiment one.

[0072] The scheme processing and analysis module is used for screening the preselected schemes according to the behavior sequence features in the user portrait to obtain primary selected schemes, screening all purchase schemes according to sandstone types to obtain to-be-analyzed schemes, determining associated items according to the user behavior sequence features, screening the primary selected schemes according to the associated items as a standard, comprehensively calculating effect indexes of the associated item schemes from accuracy, conversion rate, and user satisfaction, and transmitting the associated item schemes to the scheme processing and analysis module. The specific processing manner is the same as the processing process of step S3 in embodiment one.

[0073] The scheme processing and analyzing module is configured to sort according to the effect index, analyze the periodic purchase of the user, and generate a purchase invariable signal or a purchase change signal. For the purchase invariable signal, the scheme with the maximum effect index meeting the user area is screened to generate recommendation information. For the purchase change signal, the specific change is determined to generate an increase or decrease analysis signal.

[0074] The increase analysis signal is processed, and the related item scheme is screened according to the maximum two-effect index to generate recommendation information.

[0075] The decrease analysis signal is processed, the behavior sequence feature corresponding to the most browsing times is obtained, and the to-be-analyzed scheme is screened in combination with the user area and the purchased record to generate recommendation information. The generated recommendation information is transmitted to the recommendation information display module, and the specific processing manner is the same as the processing process of step S4 in the first embodiment.

[0076] The recommendation information display module is configured to display the obtained recommendation information to the corresponding user.

[0077] Part of the data in the above formula is calculated by taking the numerical value, and is not calculated by substituting the parameter unit. The contents not described in detail in the specification all belong to the prior art known to those skilled in the art.

[0078] The above embodiments are only used to illustrate the technical method of the present application and are not limited. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical method of the present application.

Claims

1. A sand and gravel information personalized recommendation method based on user transaction behavior, characterized by, The method specifically comprises the following steps: Collect user transaction information and perform data cleaning and desensitization to obtain preprocessed data, while establishing a user portrait and extracting time characteristics, preference characteristics and behavior sequence characteristics; Screen similar users and corresponding procurement plans based on the user portrait, and screen the procurement plans based on the extracted characteristics to obtain preselected plans; Based on the sandstone type corresponding to the behavior sequence characteristics in the user portrait, the preselected plans are screened again to obtain initial selected plans, and based on the non-purchased records, the procurement plans are screened to obtain to-be-analyzed plans; Analyze the associated items corresponding to the purchased records of the user, and screen the initial selected plans based on the associated items to obtain associated item plans, while analyzing the accuracy, conversion rate and user satisfaction indicators, and comprehensively calculating the effect indicators; According to the effect indicators, analyze the user cycle procurement situation to generate procurement unchanged signals or procurement change signals, screen the maximum effect indicator plan that meets the user area for the procurement unchanged signals to generate recommendation information; For the procurement change signals, determine the specific change situation, generate an increase or decrease analysis signal, and generate recommendation information respectively. 2.The sand information personalized recommendation method based on user transaction behavior according to claim 1, characterized in that, The specific way of screening the procurement plans to obtain preselected plans is: Obtain the user portrait and corresponding characteristics, and mark the users with an intersection of a group of characteristics as similar users, while obtaining the corresponding procurement plans as i, and i=1, 2, …, j, where j represents the number of procurement plans, and establishing the sandstone label of the procurement plan i, screening the procurement plan i based on the characteristics corresponding to the user portrait to obtain preselected plans and label them as a, and a=1, 2, …, b, where b represents the number of preselected plans. 3.The method of claim 1, wherein, The specific way of obtaining to-be-analyzed plans is: Obtain the corresponding behavior sequence characteristics in the user portrait, and obtain the purchased records and non-purchased records in the behavior sequence characteristics, while screening the preselected plans a again according to the sandstone type corresponding to the purchased records to obtain initial selected plans, then determine the sandstone type corresponding to the non-purchased records as to-be-analyzed type, and re-screen all procurement plans according to the to-be-analyzed type to obtain to-be-analyzed plans n, and n=1, 2, …, m, where m represents the number of to-be-analyzed plans. 4.The method of claim 1, wherein, The specific way of comprehensively calculating the effect indicators is: Obtain the initial selected plans and the purchased records of the user, and determine the associated items according to the purchased records, and the associated items represent the items that exist in combination, screen the initial selected plans based on the associated items to obtain associated item plans, then analyze the corresponding quantitative indicators from the accuracy, conversion rate and user satisfaction; According to the formula effect indicator=accuracy indicator×first weight+conversion rate indicator×second weight+satisfaction indicator×third weight, the effect indicators of the associated item plans are calculated. 5.The method of claim 4, wherein, The way of analyzing the corresponding quantitative indicators from the accuracy, conversion rate and user satisfaction is: An accuracy rate index is calculated according to a formula: accuracy rate = matching proportion * z1 + repeat purchase proportion * z2, wherein z1 and z2 are corresponding weight ratios; A conversion rate index is calculated according to a formula: conversion rate = click conversion rate * q1 + purchase conversion rate * q2, wherein q1 and q2 are corresponding weight ratios; A satisfaction index is calculated according to a formula: satisfaction index = score index * r1 + feedback good comment index * r2, wherein r1 and r2 are corresponding weight ratios. 6.The method of claim 1, wherein, The specific way of generating the recommendation information by screening the associated item scheme with the maximum effect index in the user area is: The associated item schemes are sorted according to the effect index from large to small, and the current user cycle purchase situation is obtained; if the cycle purchase situation is unchanged, a purchase unchanged signal is generated; otherwise, if the cycle purchase situation is changed, a purchase change analysis signal is generated; The purchase unchanged signal is analyzed, and the associated item scheme corresponding to the maximum effect index in the user area is screened as a standard to generate the recommendation information. 7.The method of claim 1, wherein, The analysis way of the purchase change signal is: The cycle purchase specific situation is obtained to generate an increase analysis signal or a decrease analysis signal; the increase analysis signal is processed, and the top two effect indexes are selected as a standard to screen the associated item schemes to generate the recommendation information; the decrease analysis signal is processed to obtain all to-be-analyzed schemes n, and the corresponding most frequently browsed sandstone type in the behavior sequence feature in the user portrait is obtained as a standard to screen the to-be-analyzed schemes n, and the user area and the purchased record are combined for comprehensive analysis to screen the to-be-analyzed schemes to generate the recommendation information.

8. A sand and gravel information personalized recommendation platform based on user transaction behavior, configured to perform the sand and gravel information personalized recommendation method based on user transaction behavior according to any one of claims 1-7, characterized in that, It comprises: a user information collection module configured to collect and pre-process user transaction information to obtain pre-processed data, and transmit the pre-processed data to a user portrait establishment module; the user portrait establishment module configured to establish a user portrait according to the pre-processed data, extract corresponding features of the user, including time features, preference features and sequence features, screen similar users according to the user portrait, screen pre-selected schemes from the purchase schemes of the similar users, and transmit the pre-selected schemes to a scheme processing and analysis module; The scheme processing analysis module is configured to filter pre-selected schemes according to the behavior sequence features in the user portrait, obtain initial selected schemes, filter all procurement schemes according to the sandstone types, obtain schemes to be analyzed, determine associated items according to the user behavior sequence features, filter the initial selected schemes according to the associated items as a standard, obtain associated item schemes, then comprehensively calculate effect indexes of the associated item schemes from accuracy, conversion rate and user satisfaction, and transmit the effect indexes to the scheme processing analysis module; The scheme processing analysis module is configured to sort the effect indexes, analyze the user periodic procurement to generate a procurement invariable signal or a procurement change signal, filter the maximum effect index schemes in line with the user area to generate recommendation information for the procurement invariable signal, determine specific change conditions for the procurement change signal, and generate an increase or decrease analysis signal; The increase analysis signal is processed, and the associated item schemes are filtered according to the maximum two effect indexes to generate recommendation information; The decrease analysis signal is processed, the sandstone type corresponding to the most browsing times of the behavior sequence features is obtained, the schemes to be analyzed are filtered according to the user area and the purchased records, and recommendation information is generated, and the generated recommendation information is transmitted to the recommendation information display module; The recommendation information display module is configured to display the obtained recommendation information to the corresponding user.

Images