Machine learning model optimization system and method based on big data analysis

Abstract

The present invention relates to the technical field of data model optimization processing, and discloses a machine learning model optimization system and method based on big data analysis, including the following steps: a user behavior data module, used to obtain the behavior data of a target user, a model building module, a model optimization module, regularly obtaining the abnormal behavior data of the target user, forming abnormal behavior node features, and reintegrating the abnormal behavior node features into a conventional machine learning model according to the weight of the abnormal behavior node features to form an optimized machine learning model. By obtaining the abnormal behavior data of the target user, new data can be continuously updated, and a new machine learning model can be established based on the abnormal behavior node features in the continuously updated data according to the abnormal behavior node features of the abnormal behavior data. The model can regularly incorporate new judgment factors to meet the growth needs of the machine learning model, which is more conducive to solving the problems caused by the changing data.

Description

Technical Field

[0001] The present invention relates to the technical field of data model optimization processing, and specifically to a machine learning model optimization system and method based on big data analysis. Background Art

[0002] Data models are used to solve data integration and prediction problems in life. Based on known data behaviors, problems can be solved quickly and easily through the integration of data models. While refining and optimizing data structures, important data parameters can be accurately displayed, thereby extracting data features with larger weights based on the above data parameters to assist users in solving problems.

[0003] As for some data features, as time accumulates, the features of some data will face an increasing phenomenon. At this time, the data model that relies on the current data features to solve the problem will lose a certain degree of accuracy. At the same time, the impact of data features is not constant over time. Inherent data models generally do not consider changes in data features. It is obviously inappropriate to use inherent data features alone to deal with data problems caused by the accumulation of changing data. Therefore, data models, also known as machine learning models, need to be continuously optimized according to changing data to ensure the growth of machine learning models. Relying on a growing learning model can solve positive problems in life more efficiently. Summary of the invention

[0004] The present invention provides a machine learning model optimization system and method based on big data analysis to promote the solution of the problems mentioned in the above-mentioned Beijing technology.

[0005] The present invention provides the following technical solution: a machine learning model optimization system based on big data analysis, comprising the following steps:

[0006] User behavior data module, used to obtain the behavior data of target users, including general behavior data and abnormal behavior data;

[0007] A model building module, used to form general behavior node features based on the general behavior data of the target user, and to build a conventional machine learning model adapted to the target user based on the weights of the general behavior node features;

[0008] The model optimization module regularly obtains the abnormal behavior data of the target users, forms abnormal behavior node features, and reintegrates the abnormal behavior node features into the conventional machine learning model according to the weights of the abnormal behavior node features to form an optimized machine learning model;

[0009] The search optimization module establishes search rules suitable for target users based on user behavior data.

[0010] As an optional solution of the machine learning model optimization system and method based on big data analysis described in the present invention, the general behavior node features are formed according to the general behavior data of the target user, specifically including:

[0011] The browsing data set of the target user is organized to form a total organized data set, and the node features of all data in the total organized data set are obtained to form the total data node features;

[0012] The target user's concerned data set is integrated to form an integrated concerned data set, the concerned integrated data set is removed from the total integrated data set to obtain the excluded integrated data set, the node features of all data in the concerned integrated data set are obtained to form the general behavior node features.

[0013] As an optional solution of the machine learning model optimization system and method based on big data analysis described in the present invention, the conventional machine learning model adapted to the target user is established according to the weight of the general behavior node features, specifically including:

[0014] Count the node features of the target user's general behavior, and count the frequency of each node feature. Node features with a frequency higher than 25% are considered high-weight node features.

[0015] Obtain all data containing high-weight node features in the exclusionary integration data set to form exclusionary data, record the exclusionary data, obtain the node features of all data in the exclusionary data to form a first node feature, remove the high-weight node features of the first node feature, and form a first exclusionary node feature;

[0016] The high-weight node features are divided into the first high-weight node feature, the second high-weight node feature and the third high-weight node feature according to the order of frequency from large to small;

[0017] To obtain target product information, it is necessary to satisfy that the target product information contains high-weight node features, and the target product must exclude the first exclusion node feature;

[0018] If the target product contains the first high-weight node feature, it is classified into the first push gradient.

[0019] If the target product only contains the second highest weight node feature, it will be classified into the second push gradient.

[0020] If the target product only contains the third highest weight node feature, it is classified into the third push gradient.

[0021] As an optional solution of the machine learning model optimization system and method based on big data analysis described in the present invention, the abnormal behavior data of the target user is regularly obtained to form abnormal behavior node features, specifically including:

[0022] Within a certain period, the behavior data of the target user is obtained again to form a changed behavior data set, and the node features of all data in the changed behavior data set are obtained to form changed data node features;

[0023] Compare the changed data node features with the total data node features to obtain a feature type different from the total data node features, and record the feature type as the target behavior node feature; name the data carrying the target behavior node feature as the abnormal behavior data;

[0024] The target user's concern data is reorganized again to form change concern behavior data, and the target behavior node features in the change concern behavior data are obtained to form new weight node features.

[0025] Eliminate the change-concern behavior data from the abnormal behavior data to obtain the change-rejection behavior data.

[0026] Integrate the data containing the new weight node feature in the change exclusion behavior data to form a second data set, obtain all node features in the second data set that are different from the new weight node feature, and form a second exclusion node feature;

[0027] The set of the second rejection node feature and the new weight node feature is the abnormal behavior node feature.

[0028] As an optional solution of the machine learning model optimization system and method based on big data analysis described in the present invention, wherein: reintegrating the weights of the abnormal behavior node features into the conventional machine learning model to form an optimized machine learning model specifically includes:

[0029] To obtain target product information, it is required that the target product information includes high-weight node features and new-weight node features, and the target product must exclude the first exclusion node features and the second exclusion node features;

[0030] If the target product contains new weight node features, it will be classified into the primary push gradient.

[0031] If the target product contains the first high-weight node feature, it is classified into the first push gradient.

[0032] If the target product only contains the second highest weight node feature, it will be classified into the second push gradient.

[0033] If the target product only contains the third highest weight node feature, it is classified into the third push gradient.

[0034] As an optional solution of the machine learning model optimization system and method based on big data analysis described in the present invention, wherein: according to the user's behavior data, establishing a search rule suitable for the target user includes: S1 obtaining the representation data of the target user;

[0035] S1 obtains the semantic search field of the target user and integrates the semantic search field and general behavior node features of the target user.

[0036] S2 classifies the semantic search fields of target users into general behavior node features.

[0037] S3 analyzes the semantic level of the target user according to the representation data of the target user, and records the relevant aggregated data corresponding to the semantic level.

[0038] S4 records the semantic floating of the target user, calculates the semantic floating amount, eliminates the semantic floating amount deviation, and identifies the real semantics for the target user;

[0039] S5 performs the retrieval process for the target user according to the target user's true semantics.

[0040] As an optional solution of the machine learning model optimization system and method based on big data analysis described in the present invention, the search optimization module establishes search rules suitable for target users based on user behavior data and also includes:

[0041] Acquire characterization data of the target user during the first period;

[0042] Obtaining characterization data of the target user during the second period and the third period;

[0043] Analyze the growth attributes of target user representation data;

[0044] Then, calculate the average growth rate, and obtain the characterization data of the target users in the current period based on the growth rate.

[0045] After obtaining the target user characterization data, execute steps S1-S5.

[0046] As an optional solution of the machine learning model optimization system and method based on big data analysis described in the present invention, wherein: the semantic degree of the target user is analyzed according to the representation data of the target user, and the relevant aggregated data corresponding to the semantic degree is recorded, specifically including:

[0047] Obtain the target user's evaluation experience of the concerned aggregated data, and divide the evaluation experience into three levels: high-quality evaluation experience, general evaluation experience, and low-quality evaluation experience.

[0048] Get the comprehensive data of attention under the high-quality evaluation experience.

[0049] Get the semantic search field of the target user,

[0050] Analyze and organize the specific data of interest expressed by the target user under the current semantic search field.

[0051] As an optional solution of the machine learning model optimization system and method based on big data analysis described in the present invention, wherein: recording the semantic floating of the target user, calculating the semantic floating amount, removing the semantic floating amount deviation, and identifying the real semantics for the target user specifically include:

[0052] Get the current representation data of the target user, denoted as Q,

[0053] Get the data representation of the specific data that the target user is concerned about under the current semantic search field, denoted as G,

[0054] Then, semantic floating amount = Q / G,

[0055] Then, the data representation amount represented by the target user's real semantics = semantic floating amount × representation data.

[0056] The present invention has the following beneficial effects:

[0057] 1. The machine learning model optimization system and method based on big data analysis can continuously update new data by acquiring the abnormal behavior data of target users, and establish a new machine learning model based on the abnormal behavior node characteristics of the abnormal behavior data and the abnormal behavior node characteristics in the continuously updated data, that is, optimize the machine learning model. The model can regularly incorporate new judgment factors to meet the growth needs of the machine learning model, which is more conducive to solving the problems caused by changing data.

[0058] 2. The machine learning model optimization system and method based on big data analysis classifies the target product containing the first high-weight node feature into the first push gradient, classifies the target product containing only the second high-weight node feature into the second push gradient, and classifies the target product containing only the third high-weight node feature into the third push gradient. Through the above method, products suitable for users can be pushed to the designated user interface more quickly and accurately, thereby speeding up the user's product selection process. At the same time, according to the high-weight node features, the product push is prioritized, making the product push service more intelligent.

[0059] 3. The machine learning model optimization system and method based on big data analysis regularly re-arranges the new data behavior of the target user, determines the second exclusion node characteristics and the new weight node characteristics, and integrates the second exclusion node characteristics and the new weight node characteristics into the conventional machine learning model, so as to enable the parameters of the learning model to be newly supplemented and form an optimized machine learning model, thereby meeting the growth needs of the machine learning model and being more conducive to solving the problems caused by changing data.

[0060] 4. The machine learning model optimization system and method based on big data analysis introduces semantic floating quantities to determine the specific product features required by the target user under the specific search semantics, thereby eliminating the deviation that may occur when pushing products, and judging whether the representation data of the target user has changed, to predict whether it is necessary to increase the representation data for product push within the current push time, thereby further improving the accuracy of product push. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a schematic block diagram of the structure of the present invention. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] Example 1

[0064] See also Figure 1 , a machine learning model optimization system based on big data analysis includes the following steps:

[0065] User behavior data module, used to obtain the behavior data of target users, including general behavior data and abnormal behavior data;

[0066] A model building module, used to form general behavior node features based on the general behavior data of the target user, and to build a conventional machine learning model suitable for the target user based on the weights of the general behavior node features;

[0067] In order to understand this solution more clearly, the behavior data in this embodiment is taken as an example of online shopping for clothes in life, and the general behavior data is the purchase record of the target user purchasing clothes;

[0068] General behavior node features can be regarded as various features that characterize the target user's purchase of clothing, such as long sleeves, short sleeves, round neck, V neck, etc.

[0069] The model optimization module regularly obtains the abnormal behavior data of the target users to form abnormal behavior node features, and reintegrates the abnormal behavior node features into the conventional machine learning model according to the weights of the abnormal behavior node features to form an optimized machine learning model;

[0070] As time goes by, new styles of clothing will emerge, and new styles can be understood as abnormal behavior data. In this embodiment, data of style products that are different from those in previous search records and that are searched by the target user through retrieval or the like are referred to as abnormal behavior data.

[0071] The search optimization module establishes search rules suitable for target users based on user behavior data.

[0072] By obtaining the abnormal behavior data of the target users, new data can be continuously updated. According to the abnormal behavior node characteristics of the abnormal behavior data, a new machine learning model can be established based on the abnormal behavior node characteristics in the continuously updated data. That is, the machine learning model is optimized. The model can regularly incorporate new judgment factors to meet the growth needs of the machine learning model, which is more conducive to solving the problems caused by changing data.

[0073] Example 2

[0074] This embodiment is a further explanation of the first embodiment, wherein, according to the general behavior data of the target user, forming the general behavior node feature specifically includes:

[0075] The browsing data set of the target user is organized to form a total organized data set, and the node features of all data in the total organized data set are obtained to form the total data node features;

[0076] The target user's concerned data set is integrated to form an integrated concerned data set, the concerned integrated data set is removed from the total integrated data set to obtain the excluded integrated data set, the node features of all data in the concerned integrated data set are obtained to form the general behavior node features.

[0077] The total data set refers to all the data information about clothing that the target user has browsed.

[0078] The focus is on the integration of data sets, which refers to all the data information about clothing that the target user has purchased.

[0079] The exclusive aggregated data set refers to all the data information about clothing that the target user has browsed but not purchased.

[0080] According to the weights of general behavior node features, a conventional machine learning model adapted to the target user is established, specifically including:

[0081] Count the node features of the target user's general behavior, and count the frequency of each node feature. Node features with a frequency higher than 25% are considered high-weight node features.

[0082] For example, there are a total of 100 general behavior node features, the general behavior node feature of long sleeves appears 40 times, the general node feature of short sleeves appears 5 times, the general node feature of round necks appears 30 times, and the general node feature of V-neck appears 25 times. Long sleeves, round necks, and V-neck are recognized as high-weight node features.

[0083] Obtain all data containing high-weight node features in the exclusionary integration data set to form exclusionary data, record the exclusionary data, obtain the node features of all data in the exclusionary data to form a first node feature, remove the high-weight node features of the first node feature, and form a first exclusionary node feature;

[0084] For example, there are 20 clothing items in the exclusion aggregation data set that contain data records of long sleeves and V-necks. The records of these 20 clothing items constitute the exclusion data. Among the 20 clothing items, the node features also include hollowing out and pleats. Then, long sleeves, round necks, V-necks, hollowing out and pleats constitute the above-mentioned first node features, and hollowing out and pleats are the first exclusion node features.

[0085] The high-weight node features are divided into the first high-weight node feature, the second high-weight node feature and the third high-weight node feature according to the order of frequency from large to small;

[0086] To obtain target product information, it is necessary to satisfy that the target product information contains high-weight node features, and the target product must exclude the first exclusion node feature;

[0087] If the target product contains the first high-weight node feature, it is classified into the first push gradient.

[0088] If the target product only contains the second highest weight node feature, it will be classified into the second push gradient.

[0089] If the target product only contains the third highest weight node feature, it is classified into the third push gradient.

[0090] In the above embodiment, long sleeves are classified into the first push gradient, round necks are classified into the second push gradient, V necks are classified into the third push gradient, and the push priority of the first push gradient is greater than the push priority of the second push gradient, and the push priority of the second push gradient is greater than the push priority of the third push gradient.

[0091] Through the above method, products suitable for users can be pushed to the designated user interface more quickly and accurately, thereby speeding up the user's product selection process. At the same time, product push can be prioritized according to the characteristics of high-weight nodes, making the product push service more intelligent.

[0092] Example 3

[0093] This embodiment is an improved description of embodiment 2. The abnormal behavior data of the target user is obtained regularly to form abnormal behavior node features, which specifically include:

[0094] Within a certain period, the behavior data of the target user is obtained again to form a changed behavior data set, and the node features of all data in the changed behavior data set are obtained to form changed data node features;

[0095] The certain period can be set manually, and can be set as one month or three months.

[0096] Compare the changed data node features with the total data node features to obtain a feature type different from the total data node features, and record the feature type as the target behavior node feature; name the data carrying the target behavior node feature as the abnormal behavior data;

[0097] The data of style products that are different from those in the previous search records and that are searched by the target user through retrieval or other means are called abnormal behavior data. For example, three months later, the node features of the clothing browsed by the target user contain node features that are completely different from before, for example, ice silk, small floral patterns, large floral patterns, zippers, and buttons. Then the data containing node features such as ice silk, small floral patterns, large floral patterns, zippers, and buttons in the clothing are called abnormal behavior data.

[0098] The target user's concern data is reorganized again to form change concern behavior data, and the target behavior node features in the change concern behavior data are obtained to form new weight node features.

[0099] The target user's interest data is organized, that is, the target user's purchased clothing belonging to the abnormal behavior data is counted, and all the feature nodes of these clothing are obtained. For example, all node features are divided into three categories: ice silk, buttons, and small patterns, and the total number of occurrences is 100.

[0100] The frequency of each node feature is counted, and the node features with a frequency higher than 25% are recognized as new weight node features. After statistics, the ice silk node feature appears 50 times, the button node feature appears 30 times, and the small flower color node feature appears 20 times. Therefore, ice silk and buttons are new weight node features. Among the new weight node features, the push priority of the ice silk node feature is higher than that of the button node feature.

[0101] Eliminate the change-concern behavior data from the abnormal behavior data to obtain the change-rejection behavior data.

[0102] Integrate the data containing the new weight node feature in the change exclusion behavior data to form a second data set, obtain all node features in the second data set that are different from the new weight node feature, and form a second exclusion node feature;

[0103] For example, the second data set contains several node features including ice silk, large flower color, and button, among which ice silk and button are new weight node features, and large flower color is the second exclusion node feature.

[0104] The set of the second rejection node feature and the new weight node feature is the abnormal behavior node feature.

[0105] According to the weight of the abnormal behavior node features, the conventional machine learning model is reintegrated to form an optimized machine learning model, which specifically includes:

[0106] To obtain target product information, it is required that the target product information includes high-weight node features and new-weight node features, and the target product must exclude the first exclusion node features and the second exclusion node features;

[0107] If the target product contains new weight node features, it will be classified into the primary push gradient.

[0108] It should be noted that the push priority of the primary push gradient is greater than the priority of the first push gradient.

[0109] If the target product contains the first high-weight node feature, it is classified into the first push gradient.

[0110] If the target product only contains the second highest weight node feature, it will be classified into the second push gradient.

[0111] If the target product only contains the third highest weight node feature, it is classified into the third push gradient.

[0112] By regularly rearranging the newly generated data behaviors of target users, determining the second exclusion node features and the new weight node features, and integrating the second exclusion node features and the new weight node features into the conventional machine learning model, the parameters of the learning model are supplemented to form an optimized machine learning model, thereby meeting the growth needs of the machine learning model and being more conducive to solving the problems caused by changing data.

[0113] Example 4

[0114] This embodiment is a further improvement made on the basis of the embodiment 3, and is used to more accurately match the products required by the user according to the semantic habits used by the target user. Specifically, according to the user's behavior data, a search rule suitable for the target user is established, including:

[0115] S1 obtains the representation data of the target user;

[0116] In this embodiment, the characterization data is the body shape data of the target user.

[0117] S1 obtains the semantic search field of the target user and integrates the semantic search field and general behavior node features of the target user.

[0118] S2 classifies the semantic search fields of target users into general behavior node features.

[0119] For example, the general behavior node features are: round collar,

[0120] The semantic search fields of the target users are wide round collar, narrow round collar, wide round collar, etc.

[0121] S3 analyzes the semantic level of the target user according to the representation data of the target user, and records the relevant aggregated data corresponding to the semantic level.

[0122] Specifically include:

[0123] Obtain the target user's evaluation experience of the concerned aggregated data, and divide the evaluation experience into three levels: high-quality evaluation experience, general evaluation experience, and low-quality evaluation experience.

[0124] Get the comprehensive data of attention under the high-quality evaluation experience.

[0125] Get the semantic search field of the target user,

[0126] Analyze and organize the specific data of interest expressed by the target user under the current semantic search field;

[0127] For example, the semantic degree words can be light color, dark color, or dark color. By obtaining the search records containing light color, dark color, or dark color in the target user's search words, obtaining the purchase records of these search records and the evaluation information under the purchase records, it is possible to obtain whether the light color artificially selected by the target user is light blue, light gray, or white, so that when pushing to the target user, products that meet the target user's standards can be pushed out more quickly.

[0128] S4 records the semantic floating of the target user, calculates the semantic floating amount, eliminates the semantic floating amount deviation, and identifies the real semantics for the target user;

[0129] In this embodiment, the semantic float is intended to clearly obtain products with complex user body types during the product selection process. Slim fit, loose fit, etc. encountered during the product selection process can all be considered as semantic floats. The semantic float that can be targeted at the target user, that is, whether the target product is slim fit or loose fit for the target user, needs to be characterized and calculated through specific data.

[0130] Specifically include:

[0131] Get the current representation data of the target user, recorded as Q, that is, record the target user's body information and digitize it, such as height, shoulder width, weight, waist circumference, etc.

[0132] Get the data representation of the specific data that the target user is concerned about under the current semantic search field, denoted as G, that is, record the product specification information of the target product, such as the shoulder width, length, chest circumference, etc.

[0133] Then, semantic floating amount = Q / G,

[0134] Then, the data representation amount represented by the target user's real semantics = semantic floating amount × representation data;

[0135] Through the above calculation method, the specifications of clothing can be represented more digitally, and clothing suitable for the target user can be pushed to the user based on the user's representation data.

[0136] S5 performs the retrieval process for the target user according to the target user's true semantics.

[0137] More specifically,

[0138] The search optimization module establishes search rules suitable for target users based on user behavior data, including:

[0139] Acquire characterization data of the target user during the first period;

[0140] Obtaining characterization data of the target user during the second period and the third period;

[0141] The interval between each of the first period, the second period and the third period can be defined as half a year.

[0142] Analyze the growth attributes of target user representation data;

[0143] Then, calculate the average growth rate, and obtain the characterization data of the target users in the current period based on the growth rate.

[0144] The growth increment from the first period to the third period is obtained, and the average growth increment within the unit is calculated. The growth increment within the current period is calculated based on the average growth increment within the unit time, so as to deduce the characterization data.

[0145] The above method can be used to determine whether there are obvious data changes in the user's body shape. If the target user is in childhood, the average growth increments of the first, second and third periods before childhood can be analyzed to infer the characterization data of the child in the current period, and clothing of appropriate size can be pushed based on the current characterization data.

[0146] After obtaining the target user characterization data, execute steps S1-S5.

[0147] By introducing semantic floating quantity, it is used to determine the specific product features required by the target user under the specific search semantics, thereby eliminating the deviation that may occur when pushing products, and judging whether the representation data of the target user has changed, to predict whether it is necessary to increase the representation data and push the product later within the current push time, thereby further improving the accuracy of product push.

[0148] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0149] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A machine learning model optimization system based on big data analysis, characterized by: The following steps are involved: User behavior data module, used to obtain the behavior data of target users, including general behavior data and abnormal behavior data; A model building module, used to form general behavior node features based on the general behavior data of the target user, and to build a conventional machine learning model suitable for the target user based on the weights of the general behavior node features; The model optimization module regularly obtains the abnormal behavior data of the target users to form abnormal behavior node features, and reintegrates the abnormal behavior node features into the conventional machine learning model according to the weights of the abnormal behavior node features to form an optimized machine learning model; The search optimization module establishes search rules suitable for target users based on user behavior data; Based on the general behavior data of the target users, the general behavior node features are formed, including: The browsing data set of the target user is organized to form a total organized data set, and the node features of all data in the total organized data set are obtained to form the total data node features; The target user's concerned data set is integrated to form an concerned integrated data set, the concerned integrated data set is removed from the total integrated data set to obtain an excluded integrated data set, and the node features of all data in the concerned integrated data set are obtained to form a general behavior node feature; According to the weights of general behavior node features, a conventional machine learning model adapted to the target user is established, specifically including: Count the node features of the target user's general behavior, and count the frequency of each node feature. Node features with a frequency higher than 25% are considered high-weight node features. Obtain all data containing high-weight node features in the exclusionary integration data set to form exclusionary data, record the exclusionary data, obtain the node features of all data in the exclusionary data to form a first node feature, remove the high-weight node features of the first node feature, and form a first exclusionary node feature; The high-weight node features are divided into the first high-weight node feature, the second high-weight node feature and the third high-weight node feature according to the order of frequency from large to small; To obtain target product information, it is necessary to satisfy that the target product information contains high-weight node features, and the target product must exclude the first exclusion node feature; If the target product contains the first high-weight node feature, it is classified into the first push gradient. If the target product only contains the second highest weight node feature, it will be classified into the second push gradient. If the target product only contains the third highest weight node feature, it is classified into the third push gradient.

2. The method for optimizing a machine learning model system based on big data analysis according to claim 1, characterized in that: Regularly obtain the target user's abnormal behavior data to form abnormal behavior node features, including: Within a certain period, the behavior data of the target user is obtained again to form a changed behavior data set, and the node features of all data in the changed behavior data set are obtained to form changed data node features; Compare the changed data node features with the total data node features to obtain a feature type different from the total data node features, and record the feature type as the target behavior node feature; The data carrying the characteristics of the target behavior node is named as abnormal behavior data; The target user's concern data is reorganized again to form change concern behavior data, and the target behavior node features in the change concern behavior data are obtained to form new weight node features. Eliminate the change-concern behavior data from the abnormal behavior data to obtain the change-rejection behavior data. Integrate the data containing the new weight node feature in the change exclusion behavior data to form a second data set, obtain all node features in the second data set that are different from the new weight node feature, and form a second exclusion node feature; The set of the second rejection node feature and the new weight node feature is the abnormal behavior node feature.

3. The method for optimizing a machine learning model system based on big data analysis according to claim 2, characterized in that: According to the weight of the abnormal behavior node features, the conventional machine learning model is reintegrated to form an optimized machine learning model, which specifically includes: To obtain target product information, it is required that the target product information includes high-weight node features and new-weight node features, and the target product must exclude the first exclusion node features and the second exclusion node features; If the target product contains new weight node features, it will be classified into the primary push gradient. If the target product contains the first high-weight node feature, it is classified into the first push gradient. If the target product only contains the second highest weight node feature, it will be classified into the second push gradient. If the target product only contains the third highest weight node feature, it is classified into the third push gradient.

4. The method for optimizing a machine learning model system based on big data analysis according to any one of claims 1 to 3, characterized in that: According to the user's behavior data, establish the search rules suitable for the target user, including: S1 obtains the representation data of the target user; S1 obtains the semantic search field of the target user and integrates the semantic search field and general behavior node features of the target user. S2 classifies the semantic search fields of target users into general behavior node features. S3 analyzes the semantic level of the target user according to the representation data of the target user, and records the relevant aggregated data corresponding to the semantic level. S4 records the semantic floating of the target user, calculates the semantic floating amount, eliminates the semantic floating amount deviation, and identifies the real semantics for the target user; S5 performs the retrieval process for the target user according to the target user's true semantics.

5. The method for optimizing a machine learning model system based on big data analysis according to claim 4, characterized in that: The search optimization module establishes search rules suitable for target users based on user behavior data, including: Acquire characterization data of the target user during the first period; Obtaining characterization data of the target user during the second period and the third period; Analyze the growth attributes of target user representation data; Then, calculate the average growth rate, and obtain the characterization data of the target users in the current period based on the growth rate. After obtaining the target user characterization data, execute steps S1-S5.

6. The method for optimizing a machine learning model system based on big data analysis according to claim 5, characterized in that: Analyze the semantic level of the target user according to the representation data of the target user, and record the relevant integration data corresponding to the semantic level, including: Obtain the target user's evaluation experience of the concerned aggregated data, and divide the evaluation experience into three levels: high-quality evaluation experience, general evaluation experience, and low-quality evaluation experience. Get the comprehensive data of attention under the high-quality evaluation experience. Get the semantic search field of the target user, Analyze and organize the specific data of interest expressed by the target user under the current semantic search field.

7. The method for optimizing a machine learning model system based on big data analysis according to claim 6, characterized in that: Record the semantic floating of target users, calculate the semantic floating amount, eliminate the semantic floating amount deviation, and identify the real semantics for target users, including: Get the current representation data of the target user, denoted as Q, Get the data representation of the specific data that the target user is concerned about under the current semantic search field, denoted as G, Then, semantic floating amount = Q / G, Then, the data representation amount represented by the target user's real semantics = semantic floating amount × representation data.

Images