An efficient database retrieval method for AI gateways used in intelligent querying

Abstract

The application relates to an AI gateway database efficient retrieval method for intelligent inquiry, and belongs to the technical field of retrieval data analysis processing. The method first determines high-frequency keywords according to the appearance frequency of each keyword in historical data and the correlation degree with other keywords and forms a high-frequency retrieval data set, then determines the importance of the current high-frequency keyword in different time periods of a day compared with other high-frequency keywords according to the statistical characteristics and data distribution characteristics of the current high-frequency keyword in the high-frequency retrieval data set, completes the sorting of all high-frequency keywords in different time periods of a day, and completes intelligent inquiry according to the sorting result, so that in the inquiry process based on the AI gateway, more refined and efficient retrieval of the AI gateway database can be realized, thereby improving the retrieval efficiency of the AI gateway database in the inquiry process.

Description

Technical Field

[0001] This invention relates to the field of data retrieval and analysis technology, and in particular to an efficient database retrieval method for AI gateways used in intelligent querying. Background Technology

[0002] The deployment of AI gateways in service halls is a major trend in the telecommunications industry. Among these gateways, the intelligent inquiry system is a crucial functional branch. Utilizing an automated retrieval system, it provides users with accurate self-service inquiry services, reducing the workload of human customer service representatives and improving service efficiency and response speed. Therefore, the retrieval speed between the inquiry system and the gateway database is paramount. The service hall's inquiry knowledge database may contain tens of thousands of FAQs (Frequently Asked Questions), business rules, and package information. Without efficient retrieval, query latency will be high, severely impacting user experience.

[0003] Existing retrieval methods, such as the TF-IDF algorithm, rely solely on word frequency to retrieve data from the database. However, the actual inquiries in business halls are complex, and relying solely on word frequency for retrieval is inefficient. This may result in problems such as excessively long retrieval times, high query latency, and low query hit accuracy, thus affecting the user experience.

[0004] In other words, current AI gateway database data retrieval methods used for intelligent inquiry suffer from the technical problem of low retrieval efficiency. Summary of the Invention

[0005] In view of this, the present invention provides an efficient retrieval method for AI gateway databases in intelligent inquiry, so as to solve the technical problem of low retrieval efficiency of AI gateway database data in the current intelligent inquiry process.

[0006] The present invention provides an efficient database retrieval method for AI gateways used in intelligent querying, comprising:

[0007] Collect the search records of each keyword for each time period before the current time to form a historical search dataset. Determine the frequency of each keyword based on the number of times any keyword appears in the historical search dataset and the number of other keywords that appear at the same time each time. Determine the high frequency search dataset based on the frequency of each keyword from the historical search dataset, and record each keyword in the high frequency search dataset as a high frequency keyword.

[0008] The current activity level of any high-frequency keyword is determined by the difference in the frequency of occurrence of any high-frequency keyword in the high-frequency retrieval dataset between the current time and historical time. The time point of each high-frequency keyword in a day when it is retrieved is taken as the retrieval time point of each high-frequency keyword. The retrieval concentration of any high-frequency keyword is determined by the distribution clustering of any high-frequency keyword in the high-frequency retrieval dataset at its retrieval time point. The importance of any high-frequency keyword in the current time period is determined by the frequency of occurrence of any high-frequency keyword in the current time period in a preset number of time periods divided by a day, combined with the retrieval concentration and current activity level of any high-frequency keyword.

[0009] All high-frequency keywords are sorted by importance within the current time period, and the search queries for each time period are completed according to the order of the high-frequency keywords in each time period.

[0010] Furthermore, determining the frequency of any keyword based on the number of times it appears in the historical retrieval dataset and the number of other keywords that appear simultaneously with it each time includes:

[0011] In the historical retrieval dataset, the difference between the number of times the current keyword appears and the number of times the keyword with the fewest occurrences is recorded as the first difference, and the difference between the number of times the keyword with the most occurrences and the number of times the keyword with the fewest occurrences is recorded as the second difference. The ratio of the first difference to the second difference is recorded as the first high frequency value.

[0012] In the historical retrieval dataset, the difference between the mean of other keywords that appear simultaneously with any keyword in each retrieval and the mean of other keywords that appear simultaneously with the keyword with the smallest mean of other keywords in each retrieval is recorded as the third difference. The difference between the mean of other keywords that appear simultaneously with the keyword with the largest mean of other keywords in each retrieval and the mean of other keywords that appear simultaneously with the keyword with the smallest mean of other keywords in each retrieval is recorded as the fourth difference. The ratio of the third difference to the fourth difference is recorded as the second high frequency value.

[0013] The average of the first and second high-frequency values ​​is used as the frequency of any keyword.

[0014] Furthermore, the step of determining the high-frequency retrieval dataset based on the high frequency of each keyword from the historical retrieval dataset includes:

[0015] The high-frequency search dataset is obtained by removing search records of keywords whose frequency is less than a preset frequency threshold from the historical search dataset.

[0016] Furthermore, determining the current activity level of any high-frequency keyword by comparing its frequency of occurrence at the current moment with that at a historical moment in the high-frequency retrieval dataset includes:

[0017] The preset duration is divided into a preceding period and a following period of equal duration. The normalized value of the ratio of the number of times any high-frequency keyword appears in the following period to the number of times it appears in the preceding period is taken as the current activity level of any high-frequency keyword.

[0018] Furthermore, determining the retrieval concentration of any high-frequency keyword by the distribution clustering of any high-frequency keyword in the high-frequency retrieval dataset at its retrieval time point includes:

[0019] One-dimensional clustering is performed on all the search time points corresponding to the current high-frequency keywords in the high-frequency search dataset. The search concentration of the current high-frequency keywords is determined based on the mean of the interval between adjacent search time points in each one-dimensional cluster and the mean of the boundary interval between any adjacent one-dimensional clusters.

[0020] Furthermore, the retrieval concentration is:

[0021] ,

[0022] in, This represents the search concentration of the q-th high-frequency keyword. This represents the number of one-dimensional clusters after performing one-dimensional clustering on all the search time points corresponding to the q-th high-frequency keyword in the high-frequency retrieval dataset. This represents the mean of the intervals between all retrieved time points within the i-th one-dimensional cluster. This represents the distance between the boundary intervals of the j-th one-dimensional cluster and the (j+1)-th one-dimensional cluster. This indicates normalization.

[0023] Furthermore, determining the importance of any high-frequency keyword in the current time period includes:

[0024] The ratio of the number of times any high-frequency keyword appears in the current time period to the total number of times any high-frequency keyword appears in all time periods is multiplied by the retrieval concentration of any high-frequency keyword to obtain the product. The product and the mean of the current activity of any high-frequency keyword are used as the importance of any high-frequency keyword in the current time period.

[0025] Furthermore, the process of completing the search query for each time period based on the order of high-frequency keywords for each time period includes:

[0026] Extract keywords from the current customer's search query as target keywords, and compare the high-frequency keywords of the current time period with the target keywords in order to complete the current customer's search query.

[0027] The advantages of this invention compared to the prior art are:

[0028] This invention first identifies high-frequency keywords and constructs a high-frequency retrieval dataset based on the frequency of occurrence of each keyword in historical data and its correlation with other keywords. Then, it determines the importance of the current high-frequency keywords relative to other high-frequency keywords at different times of the day by analyzing the statistical characteristics and data distribution characteristics of the current high-frequency keywords in the high-frequency retrieval dataset. This completes the sorting of all high-frequency keywords at different times of the day and completes intelligent queries based on the sorting results. This enables more refined and efficient retrieval of the AI ​​gateway database during the query process based on the AI ​​gateway, thereby improving query retrieval efficiency. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a flowchart illustrating an efficient database retrieval method for intelligent querying provided in Embodiment 1 of the present invention via an AI gateway. Detailed Implementation

[0031] The overall concept of this invention is as follows:

[0032] By analyzing historical inquiry data from business locations, the importance of each inquiry keyword at different times is determined based on the statistical and distribution characteristics of the historical data. This allows for the reordering and reorganization of keywords in the database according to their importance at different times, placing keywords with higher importance at the forefront of the database to improve the retrieval speed and efficiency of the inquiry system.

[0033] To further illustrate the technical solution of the present invention, specific embodiments are described below.

[0034] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of the invention include a particular feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. Furthermore, a particular feature, structure, or characteristic in one or more embodiments may be combined in any suitable form, and the terms "comprising," "including," "having," and variations thereof mean "including, but not limited to," unless otherwise specifically emphasized.

[0035] It should be understood that the sequence number of each step in the following embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0036] Method Implementation Examples:

[0037] See Figure 1 This is a flowchart illustrating an efficient database retrieval method for intelligent querying provided in Embodiment 1 of the present invention, as shown below. Figure 1 As shown, the method may include the following steps:

[0038] S101. Collect the search records of each keyword for each time period before the current time to form a historical search dataset. Determine the frequency of each keyword based on the number of times any keyword appears in the historical search dataset and the number of other keywords that appear at the same time each time it appears. Determine the high frequency search dataset based on the frequency of each keyword from the historical search dataset, and record each keyword in the high frequency search dataset as a high frequency keyword.

[0039] Before the AI ​​gateway executes the query operation, it typically stores relevant knowledge content in a data storage format of index keywords - related content of keyword queries and uses it to respond to queries. That is, each query is based on keywords. Therefore, this embodiment obtains all historical keyword retrieval data from each terminal within a preset time period before the current moment, including keywords and the time records when the keywords were retrieved. Each keyword, when retrieved, constitutes a historical retrieval data record formed by the keyword and the retrieval time, thus forming a historical retrieval dataset within the set time period. Since the storage of historical data has an upper limit, keyword retrieval historical data within one month is selected as the historical retrieval dataset.

[0040] Considering the varying frequencies of different keywords in the historical search dataset, to improve analysis efficiency, this embodiment first obtains high-frequency query keywords based on the statistical characteristics of keyword queries in the historical search dataset. Specifically, it obtains the frequency of all keywords in the historical search dataset. The more times a keyword appears, the higher the user intent of querying that keyword, and the corresponding keyword should be set at the front end of the database to facilitate querying and improve retrieval efficiency. Thus, the frequency of all keywords is first obtained. The more times a keyword appears, the stronger its frequency; conversely, the more times a keyword appears simultaneously with other keywords, the stronger the correlation between that keyword and other keywords, the easier it is to be searched along with other keywords, and the stronger its frequency. In summary, the frequency of any keyword can be constructed:

[0041]

[0042]

[0043] in, Let q be the frequency of the keyword. This represents the frequency of the q-th keyword in the historical search dataset. The higher the frequency, the stronger the user intent for that keyword and the higher its frequency. This represents the frequency of the least frequent keyword in the historical retrieval dataset. This represents the frequency of the most frequently occurring keyword in the historical retrieval dataset. To obtain the frequency of the current keyword in the historical data within the interval [0,1] through maximum and minimum value normalization, Let be the average number of other keywords that appear simultaneously with the q-th keyword in each search in the historical retrieval dataset, where ... This represents the number of other keywords that appear simultaneously with the q-th keyword in the historical retrieval dataset when the q-th keyword is retrieved (or appears) for the i-th time. This is the average number of other keywords that appear simultaneously with the q-th keyword in the historical search dataset. The larger this value, the more other keywords appear when the q-th keyword is searched, the stronger the correlation between this keyword and other keywords, the easier it is to be searched along with other keywords, and the higher the frequency of the corresponding keyword. This refers to the keyword whose average number of other keywords appearing simultaneously with each keyword in the historical search dataset during each search has the smallest average value. In other words, the average number of other keywords appearing simultaneously with each keyword in the historical search dataset during each search is calculated sequentially, and the keyword with the smallest average value is selected as the target keyword. The average number of other keywords appearing simultaneously with the target keyword during each search is then used as the target keyword. Similarly, This represents the average number of other keywords that appear simultaneously with each keyword in the historical retrieval dataset during each retrieval, corresponding to the keyword with the highest average number of other keywords appearing simultaneously during each retrieval. To obtain the frequency of the current keyword in the historical data within the interval [0,1] by normalizing the maximum and minimum values.

[0044] This yields the frequency of all keywords in the historical retrieval dataset. Since high-frequency values ​​can distinguish high-frequency from low-frequency words in the historical retrieval dataset, placing them at opposite ends of the value range, a threshold of 0.5 is used as the preset high-frequency threshold. When the frequency of the detected keyword... If a keyword is considered a high-frequency keyword, it needs to be placed in the front part of the database for sorting to improve data retrieval efficiency. This is equivalent to removing the search records of keywords with a frequency lower than the preset frequency threshold of 0.5 from the historical search dataset, and using the remaining data to form the high-frequency search dataset. For non-high-frequency keywords, sorting would bring a large amount of redundant calculation, so they are normally stored in the back part of the database.

[0045] S102, determine the current activity of any high-frequency keyword by the difference in the frequency of occurrence of any high-frequency keyword in the high-frequency retrieval dataset between the current time and historical time, take the time point of each high-frequency keyword in a day when it is retrieved as the retrieval time point of each high-frequency keyword, determine the retrieval concentration of any high-frequency keyword by the distribution clustering of any high-frequency keyword in the high-frequency retrieval dataset at its retrieval time point, and determine the importance of any high-frequency keyword in the current time period by combining the frequency of occurrence of any high-frequency keyword in the current time period in the preset number of time periods divided by a day with the retrieval concentration and current activity of any high-frequency keyword.

[0046] The above steps only determined the high-frequency keywords based on the frequency of their occurrence and their occurrence with other keywords. However, relying solely on frequency of occurrence for sorting does not take into account the actual distribution of historical data, and cannot accurately determine the relative importance of keywords under the current business conditions. Therefore, this embodiment further analyzes the statistical characteristics and data distribution characteristics of high-frequency keywords in the high-frequency retrieval dataset to determine the more active high-frequency keywords under the current business conditions and, based on this, determines the importance of each high-frequency keyword in different time periods.

[0047] The process for obtaining the current activity level of high-frequency keywords is as follows:

[0048] Inquiries received at service center terminals may be influenced by the center's activities, which change frequently. During certain activities, users may inquire more about activity-related keywords and content. Since activities are listed by date, this section obtains the recent distribution percentage of keywords in historical data. When a keyword's historical data shows a higher recent distribution compared to the past, it indicates that the keyword is currently influenced by recent activities and appears more frequently, suggesting higher current activity for that keyword.

[0049]

[0050] in, Let q be the current activity level of the q-th high-frequency keyword. Based on a set duration of one month, we take the middle value of the set duration, 15 days, and divide the preset duration into two equal periods: a preceding period and a following period. That is, the first 15 days of the set duration are considered the preceding period, and the following 15 days are considered the following period. This represents the number of times the q-th high-frequency keyword appears in the high-frequency retrieval dataset in a later time period, i.e., the recent occurrence count. This represents the number of times the q-th high-frequency keyword appeared in the previous period of the high-frequency retrieval dataset, i.e., the number of past appearances. The more recent data entries for the current high-frequency keyword compared to the number of past data entries, the better. The larger the value, the more likely the high-frequency keyword is to experience a growth trend due to recent activities, indicating higher activity levels. This indicates linear normalization of the maximum and minimum values, used to control the magnitude.

[0051] The process of determining the importance of high-frequency keywords in different time periods based on their current activity is as follows:

[0052] It's easy to understand that the keywords used in inquiries are also related to the demographics of the visiting users. For example, there will be significant differences in the inquiries of the elderly and office workers. Furthermore, the peak activity times for different social groups vary throughout the day. Therefore, the likelihood of keywords being searched at different times of the day also varies. To further improve inquiry efficiency, we divide the business hours of the day into 24-hour periods. The reason for dividing it into 24 full periods is that the activities of social groups usually change in hourly increments. Then, we analyze the importance of each high-frequency keyword within each time period.

[0053] Although the above divides a day into 24 time periods, it's easy to understand that each occurrence or retrieval time of each high-frequency keyword in the high-frequency retrieval dataset corresponds to a specific time of day. For example, 3:20 PM can be recorded as 3:20 PM. Therefore, for ease of analysis, the time point of each high-frequency keyword at each retrieval is taken as the retrieval time point of each high-frequency keyword. Then, using the retrieval time point as the standard, one-dimensional clustering, such as one-dimensional k-means clustering, is performed on all retrieval time points corresponding to the current high-frequency keyword in the high-frequency retrieval dataset, resulting in several one-dimensional clusters. If the current high-frequency keyword exhibits different inquiry peaks due to different social groups, it has strong clustering, meaning the retrieval time points of the current high-frequency keyword in the high-frequency retrieval dataset are concentrated in a few time periods. However, if the current high-frequency keyword is a universal keyword, the clustering is weaker, and the retrieval time points of the current high-frequency keyword in the high-frequency retrieval dataset will be evenly distributed throughout the high-frequency retrieval dataset.

[0054] Therefore, after performing one-dimensional clustering on all the search time points corresponding to the current high-frequency keywords in the high-frequency search dataset, the search concentration of the current high-frequency keywords can be determined based on the mean of the interval between adjacent search time points in each one-dimensional cluster and the mean of the boundary interval between any adjacent one-dimensional clusters.

[0055] Specifically, the smaller the interval between the retrieved time points within any one-dimensional cluster (i.e., the smaller the average interval between any adjacent retrieved time points), and the larger the average boundary interval between any adjacent one-dimensional clusters, the more clustered the current high-frequency keywords are, and the more likely they are to experience different inquiry peaks at different times due to different social groups. Therefore, based on the above analysis, the retrieval concentration can be further constructed as follows:

[0056]

[0057] in, Let be the search concentration of the q-th high-frequency keyword. This represents the number of one-dimensional clusters after performing one-dimensional clustering on all the search time points corresponding to the q-th high-frequency keyword in the high-frequency retrieval dataset. This represents the mean of the intervals between all retrieved time points within the i-th one-dimensional cluster. The smaller this mean interval, the more concentrated the data within that one-dimensional cluster. To obtain the centrality of all one-dimensional clusters, the smaller the value, the more clustered the high-frequency keyword is in the high-frequency retrieval dataset, using an inverse proportional form. So that changes in its value correspond to changes in search concentration. This represents the distance between the boundary intervals of the j-th one-dimensional cluster and the (j+1)-th one-dimensional cluster, or the interval between the nearest retrieved time point in the j-th one-dimensional cluster to the (j+1)-th one-dimensional cluster and the nearest retrieved time point in the (j+1)-th one-dimensional cluster to the j-th one-dimensional cluster. This represents the mean of the boundary intervals between any adjacent one-dimensional clusters in the high-frequency retrieval dataset. The larger this mean interval, the farther apart the one-dimensional clusters are within a fixed total time (i.e., a set time period), the stronger the concentration of the retrieval time points within each one-dimensional cluster, and the higher the retrieval concentration of the q-th high-frequency keyword. This indicates normalization, such as linear normalization.

[0058] Based on the obtained search concentration, the number of data points for the current keyword in each hourly time period can be further obtained. The more data points for the current keyword in an hourly time period, and the stronger the clustering characteristic of the keyword, the stronger the importance of the keyword in that time period. Combined with the activity level of the current keyword obtained above, the importance of the current high-frequency keyword in the current time period can be calculated.

[0059]

[0060] in, To determine the importance of the q-th high-frequency keyword in the k-th time period, Let q be the current activity level of the q-th high-frequency keyword. This represents the number of times the q-th keyword appears or is retrieved within the k-th time period in the high-frequency retrieval dataset. This represents the total number of occurrences or searches of the q-th high-frequency keyword in the high-frequency retrieval dataset. The larger this percentage is, the higher the importance of the q-th high-frequency keyword in the k-th time period. The higher the current activity of the q-th high-frequency keyword and the higher its importance in the k-th time period, the higher the importance of the corresponding high-frequency keyword in the k-th time period.

[0061] S103: Sort all high-frequency keywords according to their importance in the current time period, and complete the search query for each time period according to the order of the high-frequency keywords in each time period.

[0062] After obtaining the importance of all high-frequency keywords for each time period, all high-frequency keywords are sorted according to their importance for each time period. The sorting results are then applied to the database. The high-frequency keywords in the database are reorganized according to the sorted order and placed at the beginning of the database, while the remaining non-high-frequency keywords are placed at the end of the database.

[0063] In practical applications, historical data for a set duration can be selected daily for calculation to obtain the importance of each high-frequency keyword for each time period of the day. The keywords (including high-frequency and low-frequency keywords) for each time period are then reorganized and sorted. During business operations, the keywords in the current customer's search query are extracted as target keywords. The high-frequency keywords of the current time period are compared with the target keywords in the sorted order. If no matching keyword is found after the high-frequency keyword comparison is completed, the comparison continues from the remaining low-frequency keywords until the current customer's search query is completed.

[0064] Compared to existing technologies that directly use term frequency algorithms such as TF-IDF to query high-frequency keywords, which are not suitable for complex business scenarios in business halls, this invention first determines high-frequency keywords and constructs a high-frequency retrieval dataset based on the frequency of each keyword in historical data and its correlation with other keywords. Then, it determines the importance of the current high-frequency keyword relative to other high-frequency keywords in different time periods of the day by analyzing the statistical characteristics and data distribution characteristics of the current high-frequency keyword in the high-frequency retrieval dataset. It then sorts all high-frequency keywords in different time periods of the day and completes intelligent queries based on the sorting results. This enables more refined and efficient retrieval of the AI ​​gateway database during the query process based on the AI ​​gateway, thereby improving query retrieval efficiency.

[0065] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An AI gateway database efficient retrieval method for intelligent interrogation, characterized in that, The method comprises: Collecting each search record of each keyword in a preset time period before the current time to form a historical search data set, determining the high-frequency degree of any keyword according to the number of occurrences of the keyword and the number of other keywords appearing simultaneously at each occurrence in the historical search data set, determining a high-frequency search data set from the historical search data set according to the high-frequency degree of each keyword, and recording each keyword in the high-frequency search data set as a high-frequency keyword; Determining the current activity of any high-frequency keyword in the high-frequency search data set according to the difference between the occurrence frequencies of the high-frequency keyword at the current time and at the historical time, taking the time point of each high-frequency keyword when it is searched as the search time point of the high-frequency keyword, determining the search concentration of any high-frequency keyword according to the distribution aggregation of the high-frequency keyword at the search time point of the high-frequency keyword, and determining the importance of any high-frequency keyword in the current time period according to the frequency of the high-frequency keyword in the current time period divided by a preset number of time periods in a day, in combination with the search concentration and the current activity of the high-frequency keyword; Sorting all high-frequency keywords according to the importance of each high-frequency keyword in the current time period, and completing the search inquiry of each time period according to the order of high-frequency keywords in each time period; The search concentration of any high-frequency keyword in the high-frequency search data set is determined according to the distribution aggregation of the high-frequency keyword at the search time point of the high-frequency keyword, and comprises: One-dimensional clustering is performed on all search time points corresponding to the current high-frequency keyword in the high-frequency search data set, the search concentration of the current high-frequency keyword is determined according to the mean value of the interval between adjacent search time points in each one-dimensional clustering cluster and the mean value of the boundary interval between any adjacent one-dimensional clustering clusters; The search concentration is: ， wherein, represents the search concentration of the qth high-frequency keyword, represents the number of one-dimensional clustering clusters after one-dimensional clustering of all searched time points corresponding to the qth high-frequency keyword in the high-frequency search data set, represents the mean of the intervals of all searched time points within the ith one-dimensional clustering cluster, represents the distance of the boundary interval between the jth one-dimensional clustering cluster and the j+1th one-dimensional clustering cluster, represents normalization.

2. The AI gateway database efficient retrieval method for intelligent interrogation according to claim 1, characterized in that, The high-frequency degree of any keyword in the historical search data set is determined according to the number of occurrences of the keyword and the number of other keywords appearing simultaneously at each occurrence, and comprises: In the historical search data set, the difference between the number of occurrences of the current keyword and the number of occurrences of the keyword with the least number of occurrences is recorded as a first difference value, the difference between the number of occurrences of the keyword with the most number of occurrences and the number of occurrences of the keyword with the least number of occurrences is recorded as a second difference value, and the ratio of the first difference value to the second difference value is recorded as a first high-frequency value; In the historical search data set, a third difference value is recorded as a difference value between a mean value of the number of other keywords appearing simultaneously with any keyword each time the keyword is searched in the historical search data set and a mean value of the number of other keywords appearing simultaneously with the keyword corresponding to the minimum mean value of the number of other keywords appearing simultaneously with each keyword each time the keyword is searched, a fourth difference value is recorded as a difference value between a mean value of the number of other keywords appearing simultaneously with each keyword each time the keyword is searched and the mean value of the number of other keywords appearing simultaneously with the keyword corresponding to the minimum mean value of the number of other keywords appearing simultaneously with each keyword each time the keyword is searched, and a second high-frequency value is recorded as a ratio of the third difference value to the fourth difference value; a mean value of the first high-frequency value and the second high-frequency value is taken as a high-frequency degree of any keyword.

3. The AI gateway database efficient retrieval method for intelligent interrogation according to claim 1 or 2, characterized in that, The high-frequency search data set is determined according to the high-frequency degree of each keyword from the historical search data set, and includes the following steps: search records of keywords with a high-frequency degree less than a preset high-frequency degree threshold in the historical search data set are removed to obtain the high-frequency search data set.

4. The AI gateway database efficient retrieval method for intelligent interrogation according to claim 1, characterized in that, The current activity degree of any high-frequency keyword in the high-frequency search data set is determined according to a difference between the appearance frequency of the high-frequency keyword at a current time and at a historical time, and includes the following steps: a previous time period and a next time period with equal lengths are divided from the preset time length, and a normalized value of a ratio of the number of appearances of any high-frequency keyword in the next time period to the number of appearances of the high-frequency keyword in the previous time period in the high-frequency search data set is taken as the current activity degree of the high-frequency keyword.

5. The AI gateway database efficient retrieval method for intelligent interrogation according to claim 1, wherein, The importance degree of any high-frequency keyword in a current time period is determined, and includes the following steps: a product is obtained by multiplying a ratio of the number of appearances of any high-frequency keyword in the current time period to a sum of the number of appearances of the high-frequency keyword in all time periods and a search concentration degree of the high-frequency keyword, and a mean value of the current activity degree of the high-frequency keyword is taken as the importance degree of the high-frequency keyword in the current time period.

6. The AI gateway database efficient retrieval method for intelligent interrogation according to claim 1, characterized in that, The search inquiry of each time period is completed according to an order of high-frequency keywords in each time period, and includes the following steps: keywords in a current customer search inquiry sentence are extracted as target keywords, high-frequency keywords in a time period to which a current time belongs are compared with the target keywords in turn according to the arranged order, and the current customer search inquiry is completed.

Images