An intelligent query control method based on natural language interaction and an electronic device

Abstract

The application discloses a kind of intelligent query control method and electronic equipment based on natural language interaction.The method includes the following steps: building context knowledge base, converting into context vector, and storing in vector database;Convert the input natural language question into query vector;Retrieve the k context vectors most similar to the query vector from the vector database, and determine the corresponding context information;Combine the natural language question and the retrieved context information into a prompt word sequence input into a large language model (LLM) to obtain an SQL query statement, and analyze whether the natural language question contains a control intent;Execute the SQL query statement;If the natural language question contains a control intent, analyze the query result intelligently and perform the corresponding device control operation.The application can analyze the control intent in the natural language question, and after the result is queried, the corresponding device can be controlled according to the control intent to work, improving the work efficiency.

Description

Technical Field

[0001] This invention relates to the field of data query control technology, and in particular to an intelligent query control method and electronic device based on natural language interaction. Background Technology

[0002] With the continuous development of artificial intelligence, intelligent query based on natural language interaction has become a popular interaction method and has been widely applied in various industries. Currently, intelligent query systems based on natural language interaction rely on Large Language Models (LLMs) for implementation, allowing users to ask questions in natural language, which are then automatically converted into SQL statements for querying.

[0003] In data-intensive fields such as smart animal husbandry and intelligent farming, managers, farming experts and other staff members ask questions using natural language, and the intelligent query system returns query information. When staff members need to control the corresponding equipment based on the query information, they need to input equipment control commands into a separate control system, which is cumbersome. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides an intelligent query control method and electronic device based on natural language interaction. This method can analyze the control intent in natural language questions, retrieve the query results based on the natural language questions, and control the corresponding devices to operate according to the control intent, thereby improving work efficiency.

[0005] To solve the above problems, the present invention adopts the following technical solution:

[0006] The present invention provides an intelligent query control method based on natural language interaction, comprising the following steps:

[0007] S1: Construct a context knowledge base, use a vector embedding model to convert the context information in the context knowledge base into context vectors, and store them in a vector database;

[0008] S2: The input natural language question is converted into a query vector using a vector embedding model;

[0009] S3: Retrieve the k context vectors that are most similar to the query vector from the vector database and determine their corresponding context information;

[0010] S4: Combine the natural language question with the retrieved k contextual information to form a prompt word sequence. Input the prompt word sequence into a large language model (LLM). The large language model (LLM) generates an SQL query statement and analyzes whether the natural language question contains control intent.

[0011] S5: Execute the SQL query statement to obtain the query results;

[0012] S6: If the natural language question contains control intent, then the query results will be intelligently analyzed and the corresponding device control operations will be executed.

[0013] Preferably, the context information includes database schema definition statements, business logic documents, high-quality SQL query statement examples, and device control rule information.

[0014] Preferably, the database schema definition statements include statements for defining tables, columns, data types, and data relationships; the business logic document includes unstructured text describing the business meaning, calculation logic, and business rules of the tables and columns; the high-quality SQL query example includes key-value pairs verified by experts, which include natural language questions and their corresponding SQL query statements; and the device control rule information includes condition-action mapping rules, device capability description documents, historical successful control cases, and device linkage strategies.

[0015] Preferably, step S3 includes the following steps:

[0016] Calculate the cosine similarity between the query vector and all context vectors in the vector database. Sort the calculated cosine similarities in descending order, and take the context vectors corresponding to the top k cosine similarities as the k most similar context vectors. Find the context information corresponding to these k context vectors.

[0017] Preferably, the method for intelligently analyzing the query results and executing corresponding device control operations includes the following steps:

[0018] N1: Perform anomaly detection on the query results. If an anomaly is found in the query results, proceed to step N2; otherwise, terminate.

[0019] N2: Based on the query results and control intent, construct a natural language question for querying control strategies, and use a vector embedding model to convert the natural language question for control strategies into a control strategy query vector.

[0020] N3: Retrieve the k context vectors most similar to the control strategy query vector from the vector database, determine their corresponding context information, extract device control rule information from the context information, and obtain the current state of the corresponding device;

[0021] N4: Combine the query results, the extracted device control rule information, and the current status of the corresponding device into a prompt word sequence, input it into a large language model (LLM), and the large language model (LLM) generates a structured device control instruction sequence.

[0022] N5: Controls the corresponding equipment to work according to the sequence of equipment control commands;

[0023] N6: Repeatedly execute the SQL query statement within a preset time. If the query result achieves the expected effect within the preset time, the process ends; otherwise, an alarm is issued or the corresponding device is controlled to execute an upgrade control policy.

[0024] Preferably, the method for anomaly identification of query results includes a threshold detection method, a trend judgment method, or anomaly identification method based on a machine learning model.

[0025] Preferably, the device control instruction sequence includes a device identifier, device type, control command, control parameters, execution priority, and expected effect, wherein the expected effect is used to describe the target index of the control operation.

[0026] Preferably, the intelligent query control method based on natural language interaction further includes the following steps:

[0027] S7: Extract the query intent of the input natural language question, and find the sequence pattern P with the highest support as a prefix of the query intent from the sequence pattern database;

[0028] S8: Extract other query intents from the sequence pattern P, construct corresponding natural language questions based on other query intents, execute steps S2 to S5 to query the constructed natural language questions, obtain the corresponding query results, and cache them;

[0029] S9: Display the constructed natural language question on the user interface. When the user selects and inputs the constructed natural language question, the cached corresponding query results will be displayed.

[0030] Preferably, the method for generating the sequence pattern database includes the following steps:

[0031] M1: Retrieves historical user sessions, each of which includes multiple consecutive natural language questions asked by the user within a unit of time.

[0032] M2: Extracts the query intent for each natural language question;

[0033] M3: Organize all query intents corresponding to each user session into an ordered sequence S in chronological order, and combine all ordered sequences S into a sequence database. ,

[0034] ,

[0035] ,

[0036] in, For sequence database The i-th ordered sequence in the sequence, For ordered sequences The j-th query intent, where n is the sequence database. The number of ordered sequences in the sequence, where m is the number of ordered sequences. The number of query intents in the query, 1≤i≤n, 1≤j≤m;

[0037] M4: Employs a prefix projection mode growth algorithm from a sequence database. Multiple sequence patterns P are extracted from the data, the support of each sequence pattern P is calculated, and sequence patterns P with support greater than a set threshold are combined into a sequence pattern database.

[0038] The support of sequence pattern P is the percentage of ordered sequences S containing sequence pattern P in the sequence database. The proportion of the total number of ordered sequences S in the sequence is calculated using the following formula:

[0039] ,

[0040] in, The support of sequence pattern P, Indicated in sequence database The number of ordered sequences S containing sequence pattern P.

[0041] If an ordered sequence S contains all the elements of a sequence pattern P and these elements are in the same order, then the ordered sequence S contains the sequence pattern P.

[0042] An electronic device according to the present invention includes a memory and a processor, wherein executable code is stored in the memory, and when the executable code is executed by the processor, the above-described method is performed.

[0043] The beneficial effects of the present invention are: (1) It can analyze the control intent in natural language questions, and after querying the results according to the natural language questions, it can control the corresponding equipment to work according to the control intent, thereby improving work efficiency. (2) It can proactively predict the user's subsequent query needs and display them on the user operation interface, and cache the query results, which not only improves the user experience, but also speeds up the query response. Attached Figure Description

[0044] Figure 1 This is a flowchart of Example 1. Detailed Implementation

[0045] The technical solution of the present invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0046] Example 1: This example illustrates an intelligent query control method based on natural language interaction, such as... Figure 1 As shown, it includes the following steps:

[0047] S1: Construct a context knowledge base, and use vector embedding models (such as BERT model, SentenceTransformers model) to convert the context information in the context knowledge base into context vectors and store them in a vector database.

[0048] Contextual information includes database schema definition statements, business logic documents, high-quality SQL query examples, and device control rule information.

[0049] Database schema definition statements include statements used to define tables, columns, data types, and data relationships. For example, the CREATE TABLE statement.

[0050] Business logic documents consist of unstructured text that describes the business meaning of tables and columns, calculation logic (such as the "meat-to-feed ratio" calculation formula), and business rules (such as the normal threshold range of sensors).

[0051] High-quality SQL query examples include expert-verified key-value pairs, which include natural language questions and their corresponding SQL queries.

[0052] For example, the natural language question is: "What was the average temperature of pigsty No. 3 yesterday?", and the corresponding SQL query is: "SELECT AVG(temperature) FROM pig_sensor_data WHERE pen_id='003' AND DATE(timestamp)=CURDATE()-1".

[0053] Equipment control rule information includes:

[0054] Condition-action mapping rules define the abnormal range of monitored parameters and the corresponding equipment control actions. For example: temperature > 32 degrees for 10 minutes → start the fan at 70% speed;

[0055] The device capability description document records the control interface, parameter range, and response time of each IoT device. For example, the fan model Fan-01 has an adjustable wind speed of 0-100% and a response time of <2 seconds.

[0056] Historical successful control cases record the effective control strategies and their execution results in the past. For example, in the summer of 2024, when the high temperature was high, the "fans at 80% + intermittent wet curtain" strategy was adopted, and the temperature dropped from 34 degrees to 29 degrees within 5 minutes.

[0057] The equipment linkage strategy defines the collaborative control logic between multiple devices. For example: if fighting behavior is detected, the camera will turn around, record video, reduce the lights, and pause feeding.

[0058] S2: The input natural language question is converted into a query vector using a vector embedding model.

[0059] S3: Retrieve the k most similar context vectors to the query vector from the vector database and determine their corresponding context information. The specific steps are as follows:

[0060] Calculate the cosine similarity between the query vector and all context vectors in the vector database. Sort the calculated cosine similarities in descending order, and select the context vectors corresponding to the top k cosine similarities as the k most similar context vectors. Find the context information corresponding to these k context vectors. k is a positive integer.

[0061] S4: Combine the natural language question with the retrieved k contextual information to form a prompt word sequence. Input the prompt word sequence into a large language model (LLM). The large language model (LLM) generates an SQL query statement and analyzes whether the natural language question contains control intent.

[0062] For example, a natural language question is "What is the temperature of pigsty #3? If it's too hot, help me cool it down." This question contains both query intent ("query the temperature of pigsty #3") and control intent ("control the temperature of pigsty #3 to be lowered").

[0063] S5: Execute the SQL query statement to obtain the query results.

[0064] S6: If the natural language question contains control intent, then the query results will be intelligently analyzed and the corresponding device control operations will be executed.

[0065] The method for intelligently analyzing query results and executing corresponding device control operations includes the following steps:

[0066] N1: Perform anomaly detection on the query results. If an anomaly is found in the query results, proceed to step N2; otherwise, terminate the process.

[0067] Methods for identifying anomalies in query results include threshold detection methods, trend judgment methods, or anomaly identification methods based on machine learning models.

[0068] The threshold detection method is as follows: compare the numerical data in the query results with the preset threshold range; if the data exceeds the threshold range, the query results are considered abnormal.

[0069] The trend judgment method is as follows: analyze the time series query results, calculate the slope of change, and determine whether the query results continue to deteriorate. If they continue to deteriorate, the query results are judged to be abnormal.

[0070] Anomaly detection methods based on machine learning models can include deep learning-based anomaly detection methods, clustering-based anomaly detection methods, and so on.

[0071] N2: Based on the query results and control intent, construct a natural language question for querying control strategies, and use a vector embedding model to convert the natural language question for control strategies into a control strategy query vector.

[0072] For example: Based on the query results ("the temperature of pigsty No. 3 is 33 degrees") and the control intent ("control the temperature of pigsty No. 3 to decrease"), construct the natural language question of the control strategy: "How to decrease the temperature of pigsty No. 3 when it is 33 degrees".

[0073] N3: Retrieve the k context vectors most similar to the control strategy query vector from the vector database, determine their corresponding context information, extract device control rule information from the context information, and obtain the current state of the corresponding device.

[0074] N4: Combine the query results, extracted device control rule information, and the current status of the corresponding device into a prompt word sequence, input it into a large language model (LLM), and the large language model (LLM) generates a structured device control instruction sequence.

[0075] The device control command sequence is in structured JSON format and includes the following fields:

[0076] Device Identifier: A unique identifier for the controlled Internet of Things (IoT) device;

[0077] Device type: Defines the category of the device, such as sensor, camera, actuator, etc.;

[0078] Control commands: Specific operation instructions, such as start, stop, parameter settings, etc.

[0079] Control parameters: Parameter values ​​required for command execution;

[0080] Execution priority: The execution order used when controlling multiple devices;

[0081] Expected outcome: Describes the target state of the control operation.

[0082] N5: Controls the operation of the corresponding equipment according to the sequence of equipment control commands.

[0083] N6: Repeatedly executes the SQL query statement within a preset time. If the query results achieve the expected effect within the preset time, the process ends; otherwise, an alarm is issued or the corresponding device is controlled to implement an upgraded control strategy. Upgraded control strategies include increasing control strength or activating backup devices.

[0084] In this solution, a searchable contextual knowledge base is first constructed for a specific database and IoT device system, generating a corresponding vector database. Next, upon receiving a user's natural language question, semantic similarity is used to identify the k most relevant contextual information pieces. Then, the natural language question and the retrieved contextual information are combined into a prompt word sequence, input into a large-scale language model (LLM) to generate an SQL query, and the system analyzes whether the natural language question contains control intent. Finally, the SQL query is executed, retrieving the query results from the local database. If control intent is found, the query results are intelligently analyzed, and corresponding device control operations are executed. This method can analyze control intent in natural language questions, and after retrieving results based on the natural language question, it can control the operation of corresponding devices according to the control intent, forming a complete "perception-decision-control" closed loop, thus improving work efficiency.

[0085] For example:

[0086] In the field of smart animal husbandry, during the hot summer weather, the temperature in pigsty No. 3 of a certain pig farm continued to rise. The user entered a natural language question: "What is the temperature in pigsty No. 3? If it is too hot, please help me cool it down."

[0087] Performing steps S2 to S5 of the above method yields the query result: the temperature of pigsty No. 3 is 33.2℃, and the analysis shows that the natural language question contains a control intent ("control the temperature of pigsty No. 3 to drop").

[0088] Anomaly detection was performed on the query results. Since the preset temperature threshold range for pigsty No. 3 is 28-32℃, the query result is >32℃, which is an anomaly.

[0089] A natural language question for control strategy was constructed: "How to lower the temperature of pigsty No. 3 when it is 33 degrees Celsius?" A vector embedding model was used to convert the natural language question for control strategy into a control strategy query vector.

[0090] Perform steps N3 to N4 above to generate a structured sequence of equipment control instructions. Control the corresponding equipment to work according to the sequence of equipment control instructions. Specifically: start the fan in pigsty No. 3 at 70% speed.

[0091] For the next 10 minutes, execute an SQL query every minute to check the temperature of pigsty No. 3. If the temperature of pigsty No. 3 drops below 32℃ within 10 minutes, the process ends; otherwise, issue an alarm or control the fan to increase its speed.

[0092] An electronic device according to this embodiment includes a memory and a processor. The memory stores executable code, and when the executable code is executed by the processor, the above-described method is performed.

[0093] Example 2: This example of an intelligent query control method based on natural language interaction, based on Example 1, further includes the following steps:

[0094] S7: Extract the query intent of the input natural language question, and find the sequence pattern P with the highest support as the prefix (i.e. the first element) of the query intent from the sequence pattern database;

[0095] S8: Extract other query intents from the sequence pattern P, construct corresponding natural language questions based on other query intents, execute steps S2 to S5 to query the constructed natural language questions, obtain the corresponding query results, and cache them;

[0096] S9: Display the constructed natural language question on the user interface. When the user selects and inputs the constructed natural language question, the cached corresponding query results will be displayed.

[0097] The method for generating a sequence pattern database includes the following steps:

[0098] M1: Retrieves historical user sessions, each of which includes multiple consecutive natural language questions asked by the user within a unit of time.

[0099] M2: Extracts the query intent for each natural language question;

[0100] M3: Organize all query intents corresponding to each user session into an ordered sequence S in chronological order, and combine all ordered sequences S into a sequence database. ,

[0101] ,

[0102] ,

[0103] in, For sequence database The i-th ordered sequence in the sequence, For ordered sequences The j-th query intent, where n is the sequence database. The number of ordered sequences in the sequence, where m is the number of ordered sequences. The number of query intents in the query, 1≤i≤n, 1≤j≤m;

[0104] M4: Employs a prefix projection mode growth algorithm from a sequence database. Multiple sequence patterns P are extracted from the data, the support of each sequence pattern P is calculated, and sequence patterns P with support greater than a set threshold are combined into a sequence pattern database.

[0105] The support of sequence pattern P is the percentage of ordered sequences S containing sequence pattern P in the sequence database. The proportion of the total number of ordered sequences S in the sequence is calculated using the following formula:

[0106] ,

[0107] in, The support of sequence pattern P, Indicated in sequence database The number of ordered sequences S containing sequence pattern P.

[0108] If an ordered sequence S contains all the elements of a sequence pattern P and these elements are in the same order, then the ordered sequence S contains the sequence pattern P.

[0109] For example: sequence database There are 1000 ordered sequences S in the sequence database. If the number of ordered sequences S containing the sequence pattern P = "query temperature of pigsty #3, query humidity of pigsty #3" is 350, then the support of the sequence pattern P is 350 / 1000 = 35%.

[0110] In this scheme, query intent is an abstract representation of the user's query purpose. For example, a user session includes two consecutive natural language questions from the user within 3 minutes, in chronological order: "What is the current temperature of pigsty No. 3?" and "What is the current humidity of pigsty No. 3?". The query intent corresponding to the natural language question "What is the current temperature of pigsty No. 3?" is "Query the temperature of pigsty No. 3?". The query intent corresponding to the natural language question "What is the current humidity of pigsty No. 3?" is "Query the humidity of pigsty No. 3?".

[0111] All query intents corresponding to each user session are arranged into an ordered sequence S in chronological order. For example, the ordered sequence S = "Query the temperature of pigsty No. 3, query the water consumption of pigsty No. 3, query the humidity of pigsty No. 5, query the feed intake of pigsty No. 5".

[0112] A sequence pattern P is an ordered subsequence of an ordered sequence S. If an ordered sequence S contains all the elements of a sequence pattern P in the same order (but not necessarily consecutively), then the ordered sequence S contains the sequence pattern P, and the sequence pattern P is called a subsequence of the ordered sequence S.

[0113] For example: Given a sequence pattern P1 = "Query temperature of pigsty #3, query humidity of pigsty #3", for the following ordered sequence S:

[0114] The ordered sequence S1 = "Query the temperature of pigsty No. 3, query the water consumption of pigsty No. 3, query the humidity of pigsty No. 3";

[0115] The ordered sequence S2 = "Query the temperature of pigsty No. 3, query the humidity of pigsty No. 3";

[0116] The ordered sequence S3 = "Query the humidity of pigsty No. 3, query the temperature of pigsty No. 3";

[0117] The ordered sequence S4 = "Query the temperature of pigsty No. 3, query the water consumption of pigsty No. 3";

[0118] Sequence pattern P1 is a subsequence of ordered sequences S1 and S2, but not a subsequence of ordered sequence S3 (the element order is different), and not a subsequence of ordered sequence S4 (it lacks "query humidity of pigsty No. 3"). The prefix of sequence pattern P is its first element; the prefix of sequence pattern P1 is "query temperature of pigsty No. 3".

[0119] This method can proactively predict users' subsequent query needs and display them on the user interface, while caching query results, which improves user experience and speeds up query response.

[0120] For example:

[0121] The user inputs a natural language question: "What is the temperature of pigsty #3?". The query intent is extracted as "Query the temperature of pigsty #3". The sequence pattern P, which uses this query intent as a prefix, is then retrieved from the sequence pattern database, as shown in Table 1.

[0122] Table 1

[0123] Sequence pattern P Support Query the temperature and humidity of pigsty #3 45% Check the temperature and water consumption of pigsty #3. 28% Check the temperature and feed intake of pigsty #3. 22%

[0124] Extract other query intents from sequence pattern P with 45% support, namely "query the humidity of pigsty No. 3". Based on "query the humidity of pigsty No. 3", construct the corresponding natural language question: "What is the humidity of pigsty No. 3?". Execute steps S2 to S5 to query the constructed natural language question and obtain the corresponding query result: "The humidity of pigsty No. 3 is 45%", and cache it.

[0125] The user interface displays a constructed natural language question: "What is the humidity in pigsty No. 3?". When the user selects and inputs the constructed natural language question, the cached corresponding query results are displayed.

[0126] An electronic device according to this embodiment includes a memory and a processor. The memory stores executable code, and when the executable code is executed by the processor, the above-described method is performed.

Claims

1. An intelligent query control method based on natural language interaction, characterized in that, Includes the following steps: S1: Construct a context knowledge base, use a vector embedding model to convert the context information in the context knowledge base into context vectors, and store them in a vector database; S2: The input natural language question is converted into a query vector using a vector embedding model; S3: Retrieve the k context vectors that are most similar to the query vector from the vector database and determine their corresponding context information; S4: Combine the natural language question with the retrieved k contextual information to form a prompt word sequence. Input the prompt word sequence into a large language model (LLM). The large language model (LLM) generates an SQL query statement and analyzes whether the natural language question contains control intent. S5: Execute the SQL query statement to obtain the query results; S6: If the natural language question contains control intent, then perform intelligent analysis on the query results and execute the corresponding device control operations; The method for intelligently analyzing query results and executing corresponding device control operations includes the following steps: N1: Perform anomaly detection on the query results. If an anomaly is found in the query results, proceed to step N2; otherwise, terminate. N2: Based on the query results and control intent, construct a natural language question for querying control strategies, and use a vector embedding model to convert the natural language question for control strategies into a control strategy query vector. N3: Retrieve the k context vectors most similar to the control strategy query vector from the vector database, determine their corresponding context information, extract device control rule information from the context information, and obtain the current state of the corresponding device; N4: Combine the query results, the extracted device control rule information, and the current status of the corresponding device into a prompt word sequence, input it into a large language model (LLM), and the large language model (LLM) generates a structured device control instruction sequence. N5: Controls the corresponding equipment to work according to the sequence of equipment control commands; N6: Repeatedly execute the SQL query statement within a preset time. If the query result achieves the expected effect within the preset time, the process ends; otherwise, an alarm is issued or the corresponding device is controlled to execute an upgrade control policy.

2. The intelligent query control method based on natural language interaction according to claim 1, characterized in that, The context information includes database schema definition statements, business logic documents, high-quality SQL query examples, and device control rule information.

3. The intelligent query control method based on natural language interaction according to claim 2, characterized in that, The database schema definition statements include statements for defining tables, columns, data types, and data relationships; the business logic documents include unstructured text describing the business meaning, calculation logic, and business rules of the tables and columns; the high-quality SQL query examples include key-value pairs verified by experts, which include natural language questions and their corresponding SQL query statements; the equipment control rule information includes condition-action mapping rules, equipment capability description documents, historical successful control cases, and equipment linkage strategies.

4. The intelligent query control method based on natural language interaction according to claim 1, characterized in that, Step S3 includes the following steps: Calculate the cosine similarity between the query vector and all context vectors in the vector database. Sort the calculated cosine similarities in descending order, and take the context vectors corresponding to the top k cosine similarities as the k most similar context vectors. Find the context information corresponding to these k context vectors.

5. The intelligent query control method based on natural language interaction according to claim 1, characterized in that, The methods for anomaly identification of query results include threshold detection methods, trend judgment methods, or anomaly identification methods based on machine learning models.

6. The intelligent query control method based on natural language interaction according to claim 1, characterized in that, The device control instruction sequence includes device identifier, device type, control command, control parameters, execution priority, and expected effect, wherein the expected effect is used to describe the target indicator of the control operation.

7. The intelligent query control method based on natural language interaction according to claim 1, characterized in that, It also includes the following steps: S7: Extract the query intent of the input natural language question, and find the sequence pattern P with the highest support as a prefix of the query intent from the sequence pattern database; S8: Extract other query intents from the sequence pattern P, construct corresponding natural language questions based on other query intents, execute steps S2 to S5 to query the constructed natural language questions, obtain the corresponding query results, and cache them; S9: Display the constructed natural language question on the user interface. When the user selects the constructed natural language question to input, the cached corresponding query results will be displayed.

8. The intelligent query control method based on natural language interaction according to claim 7, characterized in that, The method for generating the sequence pattern database includes the following steps: M1: Retrieves historical user sessions, each of which includes multiple consecutive natural language questions asked by the user within a unit of time. M2: Extracts the query intent for each natural language question; M3: Organize all query intents corresponding to each user session into an ordered sequence S in chronological order, and combine all ordered sequences S into a sequence database. , ， ， in, For sequence database The i-th ordered sequence in the sequence, For ordered sequences The j-th query intent, where n is the sequence database. The number of ordered sequences in the sequence, where m is the number of ordered sequences. The number of query intents in the query, 1≤i≤n, 1≤j≤m; M4: Employs a prefix projection mode growth algorithm from a sequence database. Multiple sequence patterns P are extracted from the data, the support of each sequence pattern P is calculated, and sequence patterns P with support greater than a set threshold are combined into a sequence pattern database.

9. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores executable code, and when the executable code is executed by the processor, it performs the method as described in any one of claims 1-8.

Images