Product registration information management method and system based on artificial intelligence

Abstract

The present invention discloses a product registration information management method and system based on artificial intelligence, and belongs to the field of artificial intelligence technology. The method includes the following steps: extracting keywords from the registered information in the database, and classifying the information according to the fields to which the keywords belong; for the information in the same field, extracting and analyzing the features of the daily information on a daily basis to determine the common content; calculating the gradient change of the common content of the daily information, and determining the different time periods to which the information belongs; for the information in each time period, determining its key content and establishing a review template; comparing the information to be registered with the review template, obtaining review suggestions and providing feedback; predicting the template change time according to the different time periods to which the information belongs, and providing feedback reminders to the user when the application time of the information to be registered is after the predicted time. The method can effectively improve the situation that the existing technology lacks the ability to dynamically adapt to information changes and trends.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based product registration information management method and system. Background Art

[0002] In the field of product registration information management, existing background technologies mainly involve manual and semi-automatic management methods.

[0003] Manual management methods mainly rely on manpower to organize, classify and review data. Although this method has certain flexibility and adaptability, manual data processing is time-consuming and inefficient, especially when the amount of data is large, delays are prone to occur; semi-automatic management methods use certain software tools and technologies to assist in data management, such as using text processing tools for simple text analysis and classification. This method is an improvement over manual management, but it still lacks the ability to dynamically adapt to data changes and trends, and is difficult to update and adjust in a timely manner; some methods manage and review data based on predefined rules and templates, such as using keyword matching and simple rule judgment; this method improves efficiency to a certain extent, but rule-based methods have difficulty handling complex data relationships and dependencies, limiting their scope of application. Summary of the invention

[0004] The purpose of the present invention is to provide a product registration information management method and system based on artificial intelligence to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solution: a product registration information management method based on artificial intelligence, the method comprising the following steps:

[0006] S1. Extract keywords from the registered materials in the database and classify the materials according to the fields to which the keywords belong;

[0007] S2. For data in the same field, extract and analyze the features of daily data to determine the common content;

[0008] S3, calculate the gradient change of the common content of daily data and determine the different time periods to which the data belongs;

[0009] S4. For each period of data, determine its key content and establish a review template;

[0010] S5. Compare the registration materials with the review template, draw review suggestions and provide feedback;

[0011] S6. Predict the template change time according to the different time periods to which the information belongs, and provide feedback reminders to the user when the application time of the information to be registered is after the predicted time.

[0012] In step S1, the specific method of extracting keywords is:

[0013] S2-1. Remove punctuation marks and special symbols from the text information;

[0014] S2-2, use word segmentation tools to break down the text into individual words or phrases;

[0015] S2-3. Define a stop word list to remove common but meaningless words in the text;

[0016] S2-4. Count the number of times each word appears in the document TF:

[0017] TF(t,d)=f t,d / N d

[0018] Among them, TF(t,d) is the frequency of word t in data d; f t,d is the number of times word t appears in data d; N d is the total number of words in data d after removing stop words;

[0019] The word with the largest frequency TF(t,d) value is the keyword;

[0020] Define the field to which the keywords belong;

[0021] The extracted keywords are matched with the definition keywords and the materials are classified according to the fields to which the keywords belong.

[0022] In step S2, the feature extraction refers to the extraction of data content features. The content features contained in all daily data are common content features, which are represented by: F1, F2, ..., F n ; The common content feature value is represented by a format value, and each common content feature is represented by multiple formats; the common content feature value can be expressed as: V(F i )=(G i1 ,G i2 ,…,G im );

[0023] Where n is a positive integer, indicating the number of common content features; F i represents the i-th common content feature; 1≤i≤n, i is a positive integer; V(F i ) represents the common content feature F i The value of; m is a positive integer, indicating the number of formats of the i-th common content feature; G im represents the mth format value of the ith common content feature;

[0024] The common content feature values ​​are digitized and vectorized; the cosine similarity between the same common content feature values ​​of different data is calculated; all common content features and formats corresponding to a cosine similarity of 1 are the common content.

[0025] In step S3, the specific method of calculating the gradient change of the common content of the daily data is:

[0026] The common content is represented as C. As can be seen from step S2, C is composed of the common content features and formats corresponding to the cosine similarity of 1 between the common content feature values ​​of all different data; C is numerically processed; the common content of different dates is represented as T j (C), the gradient change between common contents on different dates is expressed as: ΔC = T j (C)-T j-1 (C);

[0027] Among them, T j (C) represents the common content on date j; T j-1 (C) indicates the common content as of date j-1;

[0028] The specific method for determining the different time periods to which the data belongs is as follows:

[0029] Set the threshold A1. When ΔC>A1, the gradient changes greatly. Record all dates that satisfy ΔC>A1 and arrange them in order. Divide different time periods according to the dates.

[0030] In step S4, according to the time periods divided in step S3, the key contents of the data in different time periods are determined;

[0031] The content features contained in all the data in each period are extracted as key content features, which are expressed as: K1, K2, ..., K p ; The key content feature value is represented by a format value, and each key content feature is represented by multiple formats; The key content feature value can be expressed as: V(K r )=(L r1 ,L r2 ,…,L rq );

[0032] Where p is a positive integer, indicating the number of key content features; K r represents the rth key content feature; 1≤r≤p, r is a positive integer; V(K r ) represents the key content feature K r value; q is a positive integer, indicating the number of formats of the rth key content feature; L rq represents the qth format value of the rth key content feature;

[0033] Numerize and vectorize the key content feature values; calculate the cosine similarity between the same key content feature values ​​of different materials; the key content features and formats corresponding to all cosine similarities of 1 are the key content; establish a review template based on the key content features and formats.

[0034] In step S5, the content features of the data to be registered and the corresponding format values ​​are extracted;

[0035] The comparison includes integrity comparison and format comparison;

[0036] Completeness comparison: Determine whether the content features of the material to be registered completely contain the key features in step S4 by text matching; if they completely contain, perform format comparison; if they do not completely contain, provide feedback on incomplete suggestions and corresponding modification methods;

[0037] Format comparison: After integrity comparison, select the content feature values ​​of the material to be registered and the corresponding key content feature values ​​in the template, perform numerical vectorization processing on the two under the same rule, and then calculate the cosine similarity of the two; when the cosine similarity values ​​of all the content feature values ​​of the material to be registered and the corresponding key content feature values ​​in the template are 1, feedback is given as approval suggestion, otherwise feedback is given as format error suggestion and corresponding error details and modification plan.

[0038] In step S6, the template change time is predicted based on all dates recorded in step S3 that satisfy ΔC>A1 and are arranged in order;

[0039] Data processing: Convert the time series data into a supervised learning problem; set the sliding window size U, divide the time series data into windows of size U, each window contains U days of data; the data of the first U-1 days of each window is used as input, and the data of the Uth day is used as output, forming an input-output pair for supervised learning;

[0040] Feature selection: Use year, month, and day as date features, and encode date features as [year, month, day]; use the number of days since the last gradient change as the change feature, which is the number of days since the last change; calculate the frequency of gradient changes in the time window as a statistical feature, which is the number of changes in the common content gradient in the sliding window U;

[0041] Model training:

[0042] Construct an LSTM model; the input layer accepts a time series input with a window size of U, X = {x1, x2, …, x U}; where X is the input, x represents the feature vector, which is represented by date, change and statistical features;

[0043] The LSTM layer captures the long-term dependencies of the time series, hs =σ(W h ×[h s-1 ,x s ]+b h ), where s represents the time step, s∈{1,2,…,U}, and h s is the hidden state at step s, W h is the weight matrix, x s is the input of step s, b h is the bias vector, σ is the activation function;

[0044] The fully connected layer maps the output of the LSTM layer to the predicted value, y = W y ×h s +b y ; Among them, W y is the weight matrix, b y is the bias vector;

[0045] The output layer predicts the remaining days from the current time point to the next change, Y = {y}; where Y is the output and y represents the predicted remaining days;

[0046] Calculate the time when the template changes; when the application time of the information to be registered is after the predicted template change time, provide feedback reminder to the user.

[0047] A product registration information management system based on artificial intelligence, which includes a field division module, a time period division module, a comparison feedback module and a template change prediction module;

[0048] The domain division module is used to divide the different domains to which the registered data in the database belong; the time period division module is used to divide the different time periods to which the data in different domains belong; the comparison and feedback module is used to establish templates according to the content of the data in different time periods and compare and provide feedback suggestions with the data to be registered; the template change prediction module is used to predict the template change time according to the time period information and remind the application time of the data to be registered comprehensively;

[0049] The output end of the domain division module is connected to the input end of the time period division module; the output end of the time period division module is connected to the input end of the contrast feedback module; the output end of the contrast feedback module is connected to the input end of the template change prediction module.

[0050] The domain division module includes a database unit, a keyword extraction unit and a data classification unit;

[0051] The database unit is used to provide the registered and approved data required for classification; the keyword extraction unit is used to extract the keywords of the registered and approved data; the data classification unit is used to classify the data according to the extracted keywords and divide them into different fields;

[0052] The output end of the database unit is connected to the input end of the keyword extraction unit; the output end of the keyword extraction unit is connected to the input end of the data classification unit; the output end of the data classification unit is connected to the input end of the time period division module;

[0053] The time period division module includes a feature extraction unit, a common content analysis unit, a gradient calculation unit and a time period division unit;

[0054] The feature extraction unit is used to extract common features of data in the same field; the common content analysis unit is used to determine common content based on the common features; the gradient calculation unit is used to calculate the gradient of common content on different dates; the time period division unit is used to divide the time period to which the data belongs according to the gradient change;

[0055] The output end of the feature extraction unit is connected to the input end of the common content analysis unit; the output end of the common content analysis unit is connected to the input end of the gradient calculation unit; the output end of the gradient calculation unit is connected to the input end of the time period division unit; the output end of the time period division unit is connected to the input end of the comparison feedback module.

[0056] The comparison and feedback module includes a key content analysis unit, a template establishment unit, a comparison unit and a feedback unit;

[0057] The key content analysis unit is used to determine the key content based on the data of the same period; the template establishment unit is used to establish the review template based on the key content; the comparison unit is used to compare and review the registration data based on the review template; the feedback unit is used to feedback the review suggestions;

[0058] The output end of the key content analysis unit is connected to the input end of the template establishment unit; the output end of the template establishment unit is connected to the input end of the comparison unit; the output end of the comparison unit is connected to the input end of the feedback unit and the template change prediction module;

[0059] The template change prediction module includes a data processing unit, a feature selection unit, a model training unit, a time prediction unit and a feedback reminder unit;

[0060] The data processing unit is used to define a window of a time series; the feature selection unit is used to select features according to the time series; the model training unit is used to train the LSTM model according to the features; the time prediction unit is used to predict the time when the template changes according to the model; the feedback reminder unit is used to provide feedback reminders to the user when the application time of the to-be-registered information is after the predicted template change time;

[0061] The output end of the data processing unit is connected to the input end of the feature selection unit; the output end of the feature selection unit is connected to the input end of the model training unit; the output end of the model training unit is connected to the input end of the time prediction unit; the output end of the time prediction unit is connected to the input end of the feedback reminder unit.

[0062] Compared with the prior art, the beneficial effects achieved by the present invention are: the gradient change calculation of the common content feature value of the present invention can dynamically divide different time periods, and timely update the review template according to the changes in the time period, thereby ensuring the system's sensitivity and response speed to changes in the content of the information; the present invention introduces an LSTM model to predict the time of template change, which can effectively predict the time point of future template changes, thereby reminding users before submitting information, thereby improving the intelligence of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0064] Figure 1 It is a schematic diagram of the steps of the product registration information management method based on artificial intelligence of the present invention;

[0065] Figure 2 It is a flow chart of the product registration information management system based on artificial intelligence of the present invention. DETAILED DESCRIPTION

[0066] 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.

[0067] See also Figure 1-Figure 2 The present invention provides a technical solution: a product registration information management method based on artificial intelligence, the method comprising the following steps:

[0068] S1. Extract keywords from the registered materials in the database and classify the materials according to the fields to which the keywords belong;

[0069] S2. For data in the same field, extract and analyze the features of daily data to determine the common content;

[0070] S3, calculate the gradient change of the common content of daily data and determine the different time periods to which the data belongs;

[0071] S4. For each period of data, determine its key content and establish a review template;

[0072] S5. Compare the registration materials with the review template, draw review suggestions and provide feedback;

[0073] S6. Predict the template change time according to the different time periods to which the information belongs, and provide feedback reminders to the user when the application time of the information to be registered is after the predicted time.

[0074] In step S1, the specific method of extracting keywords is:

[0075] S2-1. Remove punctuation marks and special symbols from the text information;

[0076] S2-2, use word segmentation tools to break down the text into individual words or phrases;

[0077] S2-3. Define a stop word list to remove common but meaningless words in the text;

[0078] S2-4. Count the number of times each word appears in the document TF:

[0079] TF(t,d)=f t,d / N d

[0080] Among them, TF(t,d) is the frequency of word t in data d; f t,d is the number of times word t appears in data d; N d is the total number of words in data d after removing stop words;

[0081] The word with the largest frequency TF(t,d) value is the keyword;

[0082] Define the field to which the keywords belong;

[0083] The extracted keywords are matched with the definition keywords and the materials are classified according to the fields to which the keywords belong.

[0084] In step S2, the feature extraction refers to the extraction of data content features. The content features contained in all daily data are common content features, which are represented by: F1, F2, ..., F n ; The common content feature value is represented by a format value, and each common content feature is represented by multiple formats; the common content feature value can be expressed as: V(F i )=(G i1 ,G i2 ,…,G im );

[0085] Where n is a positive integer, indicating the number of common content features; F i represents the i-th common content feature; 1≤i≤n, i is a positive integer; V(F i ) represents the common content feature F i The value of; m is a positive integer, indicating the number of formats of the i-th common content feature; G im represents the mth format value of the ith common content feature;

[0086] The common content feature values ​​are digitized and vectorized; the cosine similarity between the same common content feature values ​​of different data is calculated; all common content features and formats corresponding to a cosine similarity of 1 are the common content.

[0087] In step S3, the specific method of calculating the gradient change of the common content of the daily data is:

[0088] The common content is represented as C. As can be seen from step S2, C is composed of the common content features and formats corresponding to the cosine similarity of 1 between the common content feature values ​​of all different data; C is numerically processed; the common content of different dates is represented as T j (C), the gradient change between common contents on different dates is expressed as: ΔC = T j (C)-T j-1 (C);

[0089] Among them, T j (C) represents the common content on date j; T j-1 (C) indicates the common content as of date j-1;

[0090] The specific method for determining the different time periods to which the data belongs is as follows:

[0091] Set the threshold A1. When ΔC>A1, the gradient changes greatly. Record all dates that satisfy ΔC>A1 and arrange them in order. Divide different time periods according to the dates.

[0092] In step S4, according to the time periods divided in step S3, the key contents of the data in different time periods are determined;

[0093] The content features contained in all the data in each period are extracted as key content features, which are expressed as: K1, K2, ..., K p ; The key content feature value is represented by a format value, and each key content feature is represented by multiple formats; The key content feature value can be expressed as: V(K r )=(L r1 ,L r2 ,…,L rq );

[0094] Where p is a positive integer, indicating the number of key content features; K r represents the rth key content feature; 1≤r≤p, r is a positive integer; V(K r ) represents the key content feature K r value; q is a positive integer, indicating the number of formats of the rth key content feature; L rq represents the qth format value of the rth key content feature;

[0095] Numerize and vectorize the key content feature values; calculate the cosine similarity between the same key content feature values ​​of different materials; the key content features and formats corresponding to all cosine similarities of 1 are the key content; establish a review template based on the key content features and formats.

[0096] In step S5, the content features of the data to be registered and the corresponding format values ​​are extracted;

[0097] The comparison includes integrity comparison and format comparison;

[0098] Completeness comparison: Determine whether the content features of the material to be registered completely contain the key features in step S4 by text matching; if they completely contain, perform format comparison; if they do not completely contain, provide feedback on incomplete suggestions and corresponding modification methods;

[0099] Format comparison: After integrity comparison, select the content feature values ​​of the material to be registered and the corresponding key content feature values ​​in the template, perform numerical vectorization processing on the two under the same rule, and then calculate the cosine similarity of the two; when the cosine similarity values ​​of all the content feature values ​​of the material to be registered and the corresponding key content feature values ​​in the template are 1, feedback is given as approval suggestion, otherwise feedback is given as format error suggestion and corresponding error details and modification plan.

[0100] In step S6, the template change time is predicted based on all dates recorded in step S3 that satisfy ΔC>A1 and are arranged in order;

[0101] Data processing: Convert the time series data into a supervised learning problem; set the sliding window size U, divide the time series data into windows of size U, each window contains U days of data; the data of the first U-1 days of each window is used as input, and the data of the Uth day is used as output, forming an input-output pair for supervised learning;

[0102] Feature selection: Use year, month, and day as date features, and encode date features as [year, month, day]; use the number of days since the last gradient change as the change feature, which is the number of days since the last change; calculate the frequency of gradient changes in the time window as a statistical feature, which is the number of changes in the common content gradient in the sliding window U;

[0103] Model training:

[0104] Construct an LSTM model; the input layer accepts a time series input with a window size of U, X = {x1, x2, …, x U}; where X is the input, x represents the feature vector, which is represented by date, change and statistical features;

[0105] The LSTM layer captures the long-term dependencies of the time series, h s =σ(W h ×[h s-1 ,x s ]+b h ), where s represents the time step, s∈{1,2,…,U}, and h s is the hidden state at step s, W h is the weight matrix, x s is the input of step s, b h is the bias vector, σ is the activation function;

[0106] The fully connected layer maps the output of the LSTM layer to the predicted value, y = W y ×h s +b y ; Among them, W y is the weight matrix, b y is the bias vector;

[0107] The output layer predicts the remaining days from the current time point to the next change, Y = {y}; where Y is the output and y represents the predicted remaining days;

[0108] Calculate the time when the template changes; when the application time of the information to be registered is after the predicted template change time, provide feedback reminder to the user.

[0109] A product registration information management system based on artificial intelligence, which includes a field division module, a time period division module, a comparison feedback module and a template change prediction module;

[0110] The domain division module is used to divide the different domains to which the registered data in the database belong; the time period division module is used to divide the different time periods to which the data in different domains belong; the comparison and feedback module is used to establish templates according to the content of the data in different time periods and compare and provide feedback suggestions with the data to be registered; the template change prediction module is used to predict the template change time according to the time period information and remind the application time of the data to be registered comprehensively;

[0111] The output end of the domain division module is connected to the input end of the time period division module; the output end of the time period division module is connected to the input end of the contrast feedback module; the output end of the contrast feedback module is connected to the input end of the template change prediction module.

[0112] The domain division module includes a database unit, a keyword extraction unit and a data classification unit;

[0113] The database unit is used to provide the registered and approved data required for classification; the keyword extraction unit is used to extract the keywords of the registered and approved data; the data classification unit is used to classify the data according to the extracted keywords and divide them into different fields;

[0114] The output end of the database unit is connected to the input end of the keyword extraction unit; the output end of the keyword extraction unit is connected to the input end of the data classification unit; the output end of the data classification unit is connected to the input end of the time period division module;

[0115] The time period division module includes a feature extraction unit, a common content analysis unit, a gradient calculation unit and a time period division unit;

[0116] The feature extraction unit is used to extract common features of data in the same field; the common content analysis unit is used to determine common content based on the common features; the gradient calculation unit is used to calculate the gradient of common content on different dates; the time period division unit is used to divide the time period to which the data belongs according to the gradient change;

[0117] The output end of the feature extraction unit is connected to the input end of the common content analysis unit; the output end of the common content analysis unit is connected to the input end of the gradient calculation unit; the output end of the gradient calculation unit is connected to the input end of the time period division unit; the output end of the time period division unit is connected to the input end of the comparison feedback module.

[0118] The comparison and feedback module includes a key content analysis unit, a template establishment unit, a comparison unit and a feedback unit;

[0119] The key content analysis unit is used to determine the key content based on the data of the same period; the template establishment unit is used to establish the review template based on the key content; the comparison unit is used to compare and review the registration data based on the review template; the feedback unit is used to feedback the review suggestions;

[0120] The output end of the key content analysis unit is connected to the input end of the template establishment unit; the output end of the template establishment unit is connected to the input end of the comparison unit; the output end of the comparison unit is connected to the input end of the feedback unit and the template change prediction module;

[0121] The template change prediction module includes a data processing unit, a feature selection unit, a model training unit, a time prediction unit and a feedback reminder unit;

[0122] The data processing unit is used to define a window of a time series; the feature selection unit is used to select features according to the time series; the model training unit is used to train the LSTM model according to the features; the time prediction unit is used to predict the time when the template changes according to the model; the feedback reminder unit is used to provide feedback reminders to the user when the application time of the to-be-registered information is after the predicted template change time;

[0123] The output end of the data processing unit is connected to the input end of the feature selection unit; the output end of the feature selection unit is connected to the input end of the model training unit; the output end of the model training unit is connected to the input end of the time prediction unit; the output end of the time prediction unit is connected to the input end of the feedback reminder unit.

[0124] In this embodiment:

[0125] The registered data in the database are:

[0126] Document 1:

[0127] Application time: 2023-01-15;

[0128] Product name: 5G smartphone;

[0129] Product features: high-speed Internet access, HD calls, 5G network support;

[0130] Product size: 6.5-inch screen, 157.5mm×74.8mm×8.6mm;

[0131] Product weight: 150g;

[0132] Product application scope: personal communication, mobile office, multimedia entertainment;

[0133] Product introduction: This 5G smartphone supports the latest 5G network and can provide ultra-high-speed Internet experience. It uses a high-definition screen and a high-performance processor, suitable for various high-demand application scenarios;

[0134] Document 2:

[0135] Application time: 2023-01-15;

[0136] Product name: IoT smart gateway;

[0137] Product functions: equipment interconnection, remote control, data collection;

[0138] Product application scope: smart home, industrial automation, smart agriculture;

[0139] Product cost: $200

[0140] Product Usage Introduction: This IoT smart gateway can connect various smart devices to achieve interconnection between devices. Users can remotely control devices through mobile applications and obtain device data in real time. It is widely used in smart home and industrial automation fields;

[0141] Data 3:

[0142] Application time: 2023-01-15;

[0143] Product name: Multifunctional agricultural machinery;

[0144] Product functions: plowing, sowing, harvesting;

[0145] Product application scope: agricultural production, farmland management;

[0146] Product cost: $15,000

[0147] Product Introduction: This multifunctional agricultural machine integrates the functions of tilling, sowing and harvesting, greatly improving agricultural production efficiency. It adopts advanced mechanical structure design, easy to operate, and suitable for various farmland management work;

[0148] Keyword extraction: remove punctuation and special symbols from the text information; use word segmentation tools to break the text into individual words or phrases; define a stop word list to remove common but meaningless words in the text; count the number of times each word appears in the document

[0149] The words with the largest frequency TF(t,d) value are keywords; the keywords in data 1 are: network; the keywords in data 2 are: intelligence; the keywords in data 3 are: machinery;

[0150] Define the field to which the keywords belong;

[0151] Match the extracted keywords with the defined keywords and classify the materials according to the fields to which the keywords belong:

[0152] Data 1: Communication field; Data 2: Communication field; Data 3: Mechanical field.

[0153] For data in the same field, take data 1 and data 2 as examples: feature extraction refers to the extraction of data content features. The content features contained in all daily data are common content features, which are expressed as: application time, product name, product function; product application scope, product usage introduction; common content feature values ​​are represented by format values, and each common content feature is represented by multiple formats; taking product application scope as an example, the common content feature value can be expressed as: V (product application scope) = (font is Songti, color is black, ..., size is 4, poundage is 24 pounds);

[0154] The common content feature values ​​are digitized and vectorized; the cosine similarity between the same common content feature values ​​of different data is calculated; all common content features and formats corresponding to a cosine similarity of 1 are the common content.

[0155] In step S3, the specific method of calculating the gradient change of the common content of the daily data is:

[0156] The common content is represented as C. As can be seen from step S2, C is composed of the common content features and formats corresponding to the cosine similarity of 1 between the feature values ​​of the same common content of all different data; C is numerically processed; the gradient change ΔC between the common content of different dates is represented by the numerical common content of the day minus the numerical common content of the previous day;

[0157] Set the threshold A1 to 2. When ΔC>2, the gradient changes greatly. Record all dates that satisfy ΔC>2 and arrange them in order. Divide different time periods according to the dates.

[0158] In step S4, according to the time periods divided in step S3, the key contents of the data in different time periods are determined;

[0159] Data 1 and Data 2 belong to the same period. Taking Data 1 and Data 2 as an example, the content features contained in Data 1 and 2 are extracted as key content features, which are expressed as: application time, product name, product function; product application scope, product usage introduction; key content feature values ​​are represented by format values, and each key content feature is represented by multiple formats; taking product function as an example, the key content feature value can be expressed as: V (product function) = (font is Songti, color is black, ..., size is 5 font, poundage is 24 pounds);

[0160] Numerize and vectorize the key content feature values; calculate the cosine similarity between the same key content feature values ​​of different materials; the key content features and formats corresponding to all cosine similarities of 1 are the key content; establish a review template based on the key content features and formats.

[0161] In step S5, the content features of the data to be registered and the corresponding format values ​​are extracted;

[0162] The comparison includes integrity comparison and format comparison;

[0163] Completeness comparison: Determine whether the content features of the material to be registered completely contain the key features in step S4 by text matching; if they completely contain, perform format comparison; if they do not completely contain, provide feedback on incomplete suggestions and corresponding modification methods;

[0164] Format comparison: After integrity comparison, select the content feature values ​​of the material to be registered and the corresponding key content feature values ​​in the template, perform numerical vectorization processing on the two under the same rule, and then calculate the cosine similarity of the two; when the cosine similarity values ​​of all the content feature values ​​of the material to be registered and the corresponding key content feature values ​​in the template are 1, feedback is given as approval suggestion, otherwise feedback is given as format error suggestion and corresponding error details and modification plan.

[0165] In step S6, the template change time is predicted based on all dates recorded in step S3 that satisfy ΔC>2 and are arranged in order;

[0166] Data processing: Convert time series data into supervised learning problems; set the sliding window size U;

[0167] Feature selection: Use year, month, and day as date features; use the number of days since the last gradient change as the change feature; calculate the frequency of gradient changes within the time window as a statistical feature;

[0168] Model training: Build an LSTM model; the input layer accepts time series input with a window size of 15; the LSTM layer captures the long-term dependencies of the time series; the fully connected layer maps the output of the LSTM layer to the predicted value; the output layer predicts the remaining days from the current time point to the next change; calculates the time of template change;

[0169] When the application time of the information to be registered is after the predicted template change time, feedback reminder is given to the user.

[0170] 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.

[0171] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A product registration information management method based on artificial intelligence, characterized in that: The method comprises the following steps: S1. Extract keywords from the registered materials in the database and classify the materials according to the fields to which the keywords belong; S2. For data in the same field, extract and analyze the features of daily data to determine the common content; S3, calculate the gradient change of the common content of daily data and determine the different time periods to which the data belongs; S4. For each period of data, determine its key content and establish a review template; S5. Compare the registration materials with the review template, draw review suggestions and provide feedback; S6. Predict the template change time according to the different time periods to which the data belongs, and provide feedback reminders to the user when the application time of the registered data is after the predicted time; In step S2, the feature extraction refers to the extraction of data content features. The content features contained in all daily data are common content features, which are represented by: F1, F2, ..., F n ; The common content feature value is represented by a format value, and each common content feature is represented by multiple formats; the common content feature value can be expressed as: V(F i )=(G i1 ,G i2 ,…,G im ); Where n is a positive integer, indicating the number of common content features; F i represents the i-th common content feature; 1≤i≤n, i is a positive integer; V(F i ) represents the common content feature F i The value of; m is a positive integer, indicating the number of formats of the i-th common content feature; G im represents the mth format value of the ith common content feature; The common content feature values ​​are digitized and vectorized; the cosine similarity between the same common content feature values ​​of different data is calculated; all common content features and formats corresponding to the cosine similarity of 1 are the common content; In step S3, the specific method of calculating the gradient change of the common content of the daily data is as follows: the common content is represented as C. As can be seen from step S2, C is composed of the common content features and formats corresponding to the cosine similarity of 1 between the same common content feature values ​​of all different data; C is numerically processed; the common content of different dates is represented as T j (C), the gradient change between common contents on different dates is expressed as: ΔC = T j (C)-T j-1 (C); Among them, T j (C) represents the common content on date j; T j-1 (C) indicates the common content as of date j-1; The specific method for determining the different time periods to which the data belongs is as follows: Set the threshold A1. When ΔC>A1, the gradient changes greatly. Record all dates that satisfy ΔC>A1 and arrange them in order. Divide different time periods according to the dates. In step S6, the template change time is predicted based on all dates recorded in step S3 that satisfy ΔC>A1 and are arranged in order; Data processing: Convert the time series data into a supervised learning problem; set the sliding window size U, divide the time series data into windows of size U, each window contains U days of data; the data of the first U-1 days of each window is used as input, and the data of the Uth day is used as output, forming an input-output pair for supervised learning; Feature selection: Use year, month, and day as date features, and encode date features as [year, month, day]; use the number of days since the last gradient change as the change feature, which is the number of days since the last change; calculate the frequency of gradient changes in the time window as a statistical feature, which is the number of changes in the common content gradient in the sliding window U; Model training: Build an LSTM model; the input layer accepts time series input with a window size of U, X = {x1, x2, …, x U }; where X is the input, x represents the feature vector, which is represented by date, change and statistical features; The LSTM layer captures the long-term dependencies of the time series, h s =σ(W h ×[h s-1 ,x s ]+b h ), where s represents the time step, s∈{1,2,…,U}, and h s is the hidden state at step s, W h is the weight matrix, x s is the input of step s, b h is the bias vector, σ is the activation function; The fully connected layer maps the output of the LSTM layer to the predicted value, y = W y ×h s +b y ; Among them, W y is the weight matrix, b y is the bias vector; The output layer predicts the remaining days from the current time point to the next change, Y = {y}; where Y is the output and y represents the predicted remaining days; Calculate the time when the template changes; when the application time of the information to be registered is after the predicted template change time, provide feedback reminder to the user.

2. The product registration information management method based on artificial intelligence according to claim 1, characterized in that: In step S1, the specific method of extracting keywords is: S2-1. Remove punctuation marks and special symbols from the text information; S2-2, use word segmentation tools to break down the text into individual words or phrases; S2-3. Define a stop word list to remove common but meaningless words in the text; S2-4. Count the number of times each word appears in the document TF: TF(t,d)=f t,d / N d ; Among them, TF(t,d) is the frequency of word t in data d; f t,d is the number of times word t appears in data d; N d is the total number of words in data d after removing stop words; The word with the largest frequency TF(t,d) value is the keyword; Define the field to which the keywords belong; The extracted keywords are matched with the definition keywords and the materials are classified according to the fields to which the keywords belong.

3. The product registration information management method based on artificial intelligence according to claim 2 is characterized in that: In step S4, according to the time periods divided in step S3, the key contents of the data in different time periods are determined; The content features contained in all the data in each period are extracted as key content features, which are expressed as: K1, K2, ..., K p ; The key content feature value is represented by a format value, and each key content feature is represented by multiple formats; The key content feature value can be expressed as: V(K r )=(L r1 ,L r2 ,…,L rq ); Where p is a positive integer, indicating the number of key content features; K r represents the rth key content feature; 1≤r≤p, r is a positive integer; V(K r ) represents the key content feature K r value; q is a positive integer, indicating the number of formats of the rth key content feature; L rq represents the qth format value of the rth key content feature; Numerize and vectorize the key content feature values; calculate the cosine similarity between the same key content feature values ​​of different materials; the key content features and formats corresponding to all cosine similarities of 1 are the key content; establish a review template based on the key content features and formats.

4. The product registration information management method based on artificial intelligence according to claim 3 is characterized in that: In step S5, the content features of the data to be registered and the corresponding format values ​​are extracted; The comparison includes integrity comparison and format comparison; Completeness comparison: Determine whether the content features of the material to be registered completely contain the key features in step S4 by text matching; if they completely contain, perform format comparison; if they do not completely contain, provide feedback on incomplete suggestions and corresponding modification methods; Format comparison: After integrity comparison, select the content feature values ​​of the material to be registered and the corresponding key content feature values ​​in the template, perform numerical vectorization processing on the two under the same rule, and then calculate the cosine similarity of the two; when the cosine similarity values ​​of all the content feature values ​​of the material to be registered and the corresponding key content feature values ​​in the template are 1, feedback is given as approval suggestion, otherwise feedback is given as format error suggestion and corresponding error details and modification plan.

5. The product registration information management system based on artificial intelligence is characterized by: The system includes a domain division module, a time period division module, a comparison feedback module and a template change prediction module; The domain division module is used to divide the different domains to which the registered data in the database belong; the time period division module is used to divide the different time periods to which the data in different domains belong; the comparison and feedback module is used to establish templates according to the content of the data in different time periods and compare and provide feedback suggestions with the data to be registered; the template change prediction module is used to predict the template change time according to the time period information and remind the application time of the data to be registered comprehensively; The output end of the domain division module is connected to the input end of the time period division module; the output end of the time period division module is connected to the input end of the comparison feedback module; The output end of the comparison feedback module is connected to the input end of the template change prediction module; The domain division module includes a database unit, a keyword extraction unit and a data classification unit; The database unit is used to provide the registered and approved data required for classification; the keyword extraction unit is used to extract the keywords of the registered and approved data; the data classification unit is used to classify the data according to the extracted keywords and divide them into different fields; The output end of the database unit is connected to the input end of the keyword extraction unit; the output end of the keyword extraction unit is connected to the input end of the data classification unit; the output end of the data classification unit is connected to the input end of the time period division module; The time period division module includes a feature extraction unit, a common content analysis unit, a gradient calculation unit and a time period division unit; The feature extraction unit is used to extract common features of data in the same field; the common content analysis unit is used to determine common content based on the common features; the gradient calculation unit is used to calculate the gradient of common content on different dates; the time period division unit is used to divide the time period to which the data belongs according to the gradient change; The output end of the feature extraction unit is connected to the input end of the common content analysis unit; the output end of the common content analysis unit is connected to the input end of the gradient calculation unit; the output end of the gradient calculation unit is connected to the input end of the time period division unit; the output end of the time period division unit is connected to the input end of the comparison feedback module; The comparison and feedback module includes a key content analysis unit, a template establishment unit, a comparison unit and a feedback unit; The key content analysis unit is used to determine the key content based on the data of the same period; the template establishment unit is used to establish the review template based on the key content; the comparison unit is used to compare and review the registration data based on the review template; the feedback unit is used to feedback the review suggestions; The output end of the key content analysis unit is connected to the input end of the template establishment unit; the output end of the template establishment unit is connected to the input end of the comparison unit; the output end of the comparison unit is connected to the input end of the feedback unit and the template change prediction module; The template change prediction module includes a data processing unit, a feature selection unit, a model training unit, a time prediction unit and a feedback reminder unit; The data processing unit is used to define a window of a time series; the feature selection unit is used to select features according to the time series; the model training unit is used to train the LSTM model according to the features; the time prediction unit is used to predict the time when the template changes according to the model; the feedback reminder unit is used to provide feedback reminders to the user when the application time of the to-be-registered information is after the predicted template change time; The output end of the data processing unit is connected to the input end of the feature selection unit; the output end of the feature selection unit is connected to the input end of the model training unit; the output end of the model training unit is connected to the input end of the time prediction unit; the output end of the time prediction unit is connected to the input end of the feedback reminder unit.

Images