A combined retrieval optimization method and system based on artificial intelligence

Abstract

The present invention discloses a combined search optimization method and system based on artificial intelligence, which relates to the field of artificial intelligence technology and solves the problem that its artificial intelligence simply outputs the search content associated with the search text and cannot determine the extended text of such search text based on historical records. The present invention performs initial processing on the input search text, removes useless words and related symbols inside, obtains the corresponding keyword text, and based on the relevant vector of the keyword text, identifies other combined texts with such relevant vectors from the historical search records, and then confirms the vector position, vector length ratio and search times of the combined text in turn, locks the similar text closest to the search text, and synchronously searches the similar text with the search text, which can not only ensure the normal search of the search text, but also achieve the effect of combined search synchronously.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based combined retrieval optimization method and system. Background Art

[0002] Artificial intelligence technology can be used to optimize the index construction of the retrieval system; for example, natural language processing (NLP) technology can be used to perform lexical analysis, syntactic analysis, and semantic understanding of documents to extract more valuable keywords and semantic information to build indexes.

[0003] The application with announcement number CN117688136B discloses a combined retrieval optimization method and system based on artificial intelligence, wherein the combined retrieval optimization method based on artificial intelligence includes the following steps: S1: obtaining text data from multiple trusted data sources according to preset acquisition conditions; S2: analyzing the text data, determining the target category, and obtaining the storage address according to the target category; S3: processing the text data, obtaining retrieval data, and storing the retrieval data in a corresponding retrieval database according to the storage address; S4: receiving a retrieval request, analyzing the retrieval request, determining the retrieval category, and obtaining the retrieval results from the corresponding retrieval database according to the retrieval category. This application can improve retrieval efficiency and retrieval accuracy.

[0004] With the advancement of the times, its artificial intelligence can quickly find related search content based on the corresponding search text and output the content. However, in the actual application process, its artificial intelligence still has the following deficiencies that need to be improved:

[0005] 1. Its artificial intelligence simply outputs the search content associated with the search text, and cannot determine the extended text of such search text based on historical records, resulting in the inability to diversify the search content, and further resulting in the inability to meet the search needs of searchers;

[0006] 2. In the actual multi-search process, only one search can be performed one by one, and the search cannot be performed simultaneously, so the searched content cannot be guaranteed to be output synchronously. Summary of the invention

[0007] In view of the deficiencies in the prior art, the present invention provides a combined retrieval optimization method and system based on artificial intelligence, which solves the problem that its artificial intelligence simply outputs the retrieval content associated with the retrieval text and cannot determine the extended text of such retrieval text based on historical records.

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: a combined retrieval optimization method based on artificial intelligence, comprising the following steps:

[0009] Step 1: Receive and pre-process the text input by the user: first perform format standardization on the text to confirm the standard text, then remove symbols from the standard text to obtain the keyword text: based on the standard format preset in the artificial intelligence, identify whether the format of the input text is consistent with the standard format. If it is consistent, mark the input text as the standard text; if it is not consistent, perform format brush processing on the input text to adjust the input text format to the standard format, and mark the adjusted text as the standard text; based on the preset symbol database, identify the symbols and spaces in the standard text, and remove the identified special symbols and spaces to obtain the keyword text;

[0010] Step 2: reprocess the determined keyword text: check the keyword text with a preset word library, lock the exclusion words, and then remove the exclusion words in the keyword text to confirm the feature text: based on the confirmed keyword text, check the words in the keyword text with a preset word library, and mark the related words with the check results as exclusion words, based on the marked exclusion words, remove the exclusion words from the keyword text, and mark the processed keyword text as the feature text;

[0011] Step 3: Based on the confirmed feature text, lock the feature vector associated with this feature text from the preset vocabulary vector library, and then identify other combined search texts with this feature vector from the past historical search data, select similar texts from the identified several groups of combined search texts, and use the selected similar texts and feature texts as synchronous search texts. The specific sub-steps are:

[0012] S31, based on the confirmed feature text and a preset vocabulary vector library, the vocabulary vector library is a preset vector library, which stores feature vectors corresponding to different words, confirming the feature vector associated with the vocabulary in the feature text from the vocabulary vector library, and then sorting the plurality of groups of feature vectors associated with the feature text in a front-to-back manner according to the vocabulary, and confirming the feature vector set;

[0013] S32, based on the confirmed feature vector set, confirming from past historical search data a combined search text including all feature vectors in the feature vector set, wherein the multiple groups of feature vectors associated in the combined search text include all feature vectors in the feature vector set and other feature vectors;

[0014] S33, confirm the position of the feature vector set again, identify the feature vector subsequently associated with the corresponding feature vector in the feature vector set, confirm the feature vector that belongs to the same vector as the feature vector in the feature vector set from the confirmed combined search text, mark the confirmed feature vector as the main vector, hide other feature vectors in the combined search text that are not marked as the main vector, identify whether the position relationship between the multiple groups of main vectors is consistent with the position relationship of the feature vectors in the feature vector set, if consistent, mark the combined search text as the pending text, and display the feature vector hidden in the pending text, if not consistent, no calibration is performed;

[0015] S34, based on the determined pending text, identify the proportion of the total length of its main vector in the total length of the overall feature vector: mark the length of each main vector as L i-k , where i represents different pending texts, k represents different main vectors in the pending text, and the lengths of several main vectors L i-k Perform summation to confirm the first total value ZL i , and then determine the length of all feature vectors in the pending text and perform summation to confirm the second sum value BL i , using ZB i =ZL i ÷BL i Confirm the percentage value ZB associated with this pending text i , will satisfy: ZB i ≥80% of the pending texts are marked as similar texts, otherwise no marking is performed;

[0016] S35, based on the marked similar texts, identifying the number of searches for the corresponding similar texts in the historical data, and sorting the number of searches from large to small, selecting the similar texts associated with the first four groups, and using the similar texts and the determined characteristic texts as synchronous search texts;

[0017] Step 4: Based on the determined synchronous search text, perform synchronous search on different texts in the synchronous search text by executing a multi-threaded synchronous search method, and monitor the search rate in real time. Based on the analysis results of the real-time monitoring, adjust the computing power associated with different threads in real time. The specific sub-steps are:

[0018] S41, according to the number of texts in the synchronous search text, the same number of threads are executed, and based on the computing power held by the artificial intelligence, the computing power is evenly divided according to the number of threads, so that each thread can evenly share the same computing power, and each different thread searches for different texts according to the evenly divided computing power, and based on the search process, the search rate is monitored in real time;

[0019] S42, calibrate the retrieval rate monitored in real time by each different thread as SV q , where q represents different threads, and multiple sets of retrieval rates SV monitored in real time q Perform variance processing, confirm the rate variance F, and check the rate variance F with the preset value Y1:

[0020] If F>Y1, the thread associated with the current highest search rate is confirmed and marked as the high thread, and the thread associated with the current lowest search rate is marked as the low thread. The computing power is allocated from the high thread to the low thread until the associated rate variance F≤Y1 is stopped. If the search rate of the high thread is not the highest value or the search rate of the low thread is not the lowest value, the high thread or the low thread is reconfirmed, and the computing power is allocated according to the reconfirmed high thread and low thread, where the high thread and the low thread will change with the specific computing power allocation process;

[0021] If F≤Y1, determine the current moment, and confirm the search rate generated by the corresponding thread in the past moments based on the current moment, and confirm the search rate curve generated by the corresponding thread based on the timeline. From this search rate curve, determine the change value between adjacent moments, where the change value = the search rate of the next moment - the search rate of the previous moment, and select the maximum change value and the minimum change value from the confirmed groups of change values, and calculate the search rate SV associated with the current moment based on the maximum change value, the minimum change value, and the current moment. q , predict the highest search rate and the lowest search rate of the corresponding thread at the next moment, and the highest search rate = SV q + Maximum change value, minimum retrieval rate = SV q +Minimum change value;

[0022] The highest retrieval rate and the lowest retrieval rate associated with each thread at the next moment are confirmed in the same manner;

[0023] The highest or lowest search rate in each group of threads is randomly combined. When combining, each group of threads selects a group of search rates. The combined groups of search rates are processed for variance to identify whether there is a combination of F>Y1. If so, the computing power allocation method of F>Y1 is directly used to allocate computing power between threads. If not, continuous monitoring and variance determination can be performed.

[0024] S43. The text contents associated with each text search in the synchronous search text are synchronously displayed for review by the corresponding search personnel.

[0025] Preferably, a combined retrieval optimization system based on artificial intelligence comprises:

[0026] The text initial processing end receives and pre-processes the text input by the user, first performs format standardization on the text, and then removes symbols from the text that has completed format standardization to obtain keyword text;

[0027] The feature determination end reprocesses the keyword text based on the determined keyword text, checks the keyword text with a preset word library, locks in and removes words, and partially removes words in the keyword text to confirm the feature text;

[0028] The combined search text determination end, based on the confirmed feature text, locks the feature vector associated with the feature text from the preset vocabulary vector library, and then identifies other combined search texts with the feature vector from the past historical search data, selects similar texts from the identified groups of combined search texts, and uses the selected similar texts and the feature text as synchronous search texts;

[0029] The retrieval process optimization end executes a multi-threaded synchronous retrieval method based on the determined synchronous retrieval text, synchronously retrieves different texts within the synchronous retrieval text, and monitors its retrieval rate in real time. Based on the analysis results of real-time monitoring, the computing power associated with different threads is adjusted in real time.

[0030] The present invention provides a combined search optimization method and system based on artificial intelligence. Compared with the prior art, it has the following beneficial effects:

[0031] The present invention performs initial processing on the input search text, removes useless words and related symbols inside, obtains the corresponding keyword text, and based on the related vector of the keyword text, identifies other combined texts with such related vectors from the historical search records, and then sequentially confirms the vector position, vector length ratio and search times of the combined text, locks the similar text closest to the search text, and synchronously searches the similar text with the search text, which not only ensures the normal search of the search text, but also achieves the effect of combined search. Based on the input text, its artificial intelligence automatically combines the more common combined search texts to ensure its specific search effect and improve the search experience of its search personnel;

[0032] In order to achieve synchronous output and relevant comparison of the corresponding search content, a multi-threaded synchronous search method is adopted to perform real-time search on each text. During the search process, in order to ensure that the output time is relatively consistent, it is prioritized to ensure that the search rate of each thread is relatively consistent, so as to complete the corresponding text content synchronous display process, facilitate relevant search personnel to compare the content, and achieve better processing results. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a schematic diagram of the process of the present invention;

[0034] Figure 2 It is a schematic diagram of the principle framework of the present invention. DETAILED DESCRIPTION

[0035] 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.

[0036] First embodiment

[0037] See also Figure 1 , the present application provides a combined retrieval optimization method based on artificial intelligence, comprising the following steps:

[0038] Step 1: Receive and pre-process the text input by the user, first perform format standardization on the text to confirm the standard text, then remove symbols from the standard text to obtain the keyword text. During the format standardization, a standard format is set in the artificial intelligence. If the input text does not belong to such a standard format, a format brush is performed to adjust it to the standard format, and then remove the relevant symbols, spaces, etc. that appear inside to obtain the corresponding keyword text;

[0039] Among them, the sub-steps of obtaining the keyword text are:

[0040] S11. Based on the standard format preset in the artificial intelligence, identify whether the format of the input text is consistent with the standard format. If it is consistent, mark the input text as the standard text. If it is not consistent, perform format brush processing on the input text to adjust the format of the input text to the standard format, and mark the adjusted text as the standard text.

[0041] S12, based on a preset symbol database, identifying symbols and spaces in the standard text, and removing the identified special symbols and spaces to obtain keyword text (the symbols are symbols associated with daily input, and the spaces can be directly identified by artificial intelligence. The symbols are compared with the relevant symbols pre-stored in the symbol database, and it is determined that the corresponding area belongs to the symbol, and then removed);

[0042] Step 2: Based on the determined keyword text, the keyword text is reprocessed, the keyword text is checked with the preset word library, the words to be eliminated are locked, and the words in the keyword text are partially removed to confirm the feature text. Specifically, the preset word library is a related word library containing modal particles or useless words, such as: I want to find a restaurant, in which "ah", "I", "want" and so on are all specific useless words that appear in the corresponding word library, which can be directly eliminated by checking, so as to obtain the corresponding feature text: find a restaurant;

[0043] The specific sub-steps for confirming the feature text are:

[0044] Based on the confirmed keyword text, the words in the keyword text are checked with the preset word library, and the related words with the check results are marked as eliminated words. The words stored in the preset word library are all preset words, which are prepared in advance by relevant operators. Based on the marked eliminated words, the eliminated words are eliminated from the keyword text, and the processed keyword text is marked as feature text. The useless words are eliminated from the corresponding keyword text, which can effectively ensure the determination accuracy of the subsequent word vector to achieve a better combination retrieval effect;

[0045] Step 3: Based on the confirmed feature text, lock the feature vector associated with this feature text from the corresponding vocabulary vector library, and then identify other combined search texts with this feature vector from the past historical search data, select similar texts from the identified several groups of combined search texts, and use the selected similar texts and feature texts as synchronous search texts. Specifically, in some cases, because the corresponding search personnel are not good at searching, they will enter keywords they only want to know, such as "restaurant". When its artificial intelligence obtains such text, it will directly output explanations about the restaurant, etc., but the current output text is not the question the current search personnel want. Therefore, based on this type of "restaurant", its artificial intelligence combines itself and outputs based on past combined search texts, such as "nearby restaurants", "delicious restaurants", "restaurant location", "restaurant ranking", etc., to facilitate the corresponding search personnel to confirm their actual needs, and then perform specific output of the text, which can effectively improve the search experience of the corresponding search personnel and simultaneously improve the practicality of this artificial intelligence. Among them, the specific sub-steps of selecting similar texts are:

[0046] S31, based on the confirmed feature text and a preset vocabulary vector library, the vocabulary vector library is a preset vector library, which stores feature vectors corresponding to different words, confirming the feature vector associated with the vocabulary in the feature text from the vocabulary vector library, and then sorting the plurality of groups of feature vectors associated with the feature text in a front-to-back manner according to the vocabulary, and confirming the feature vector set;

[0047] S32, based on the confirmed feature vector set, confirm from the past historical search data a combined search text including all feature vectors in the feature vector set, wherein the multiple groups of feature vectors associated with the combined search text include and exceed all feature vectors in the feature vector set, that is, when the feature vector set includes vector 1, vector 2 and vector 3, the feature vectors associated with the confirmed combined search text include all three vectors and also include other vectors;

[0048] S33, confirm the position of the feature vector set again, identify the feature vector associated with the corresponding feature vector in the feature vector set, confirm the feature vector that belongs to the same vector as the feature vector in the feature vector set from the confirmed combined search text, mark the confirmed feature vector as the main vector (its main vector all appears in the feature vector set), hide other feature vectors in the combined search text that are not marked as the main vector, identify whether the position relationship between the corresponding main vectors is consistent with the position relationship of the feature vectors in the feature vector set, if consistent, mark the combined search text as the pending text, and perform a search on the hidden feature vectors in the pending text. If they are not consistent, no calibration is performed. For example, the feature vectors in the proposed feature vector set are: {vector 1, vector 2, vector 3}, and there is a set of combined search texts, and the feature vectors in the combined search texts are: {vector 1, vector 4, vector 2, vector 5, vector 3}. After confirming that the main vectors are vector 1, vector 2, and vector 3, vector 4 and vector 5 in the combined search text are hidden. After hiding, the position sorting of the feature vectors in the combined search text is consistent with the feature vectors in the feature vector set. Then this combined search text is calibrated as pending text, and the hidden vectors 4 and vector 5 are displayed.

[0049] S34, based on the determined pending text, identify the proportion of the total length of its main vector in the total length of the overall feature vector, and mark the length of each main vector as L i-k , where i represents different pending texts, k represents different main vectors in the pending text, and the lengths of several main vectors L i-k Perform summation to confirm the first total value ZL i , and then determine the length of all feature vectors in the pending text and perform summation to confirm the second sum value BL i , using ZB i =ZL i ÷BL i Confirm the percentage value ZB associated with this pending text i , will satisfy: ZB i ≥80% of the pending texts are marked as similar texts, otherwise no marking is performed (the reason for determining ZB i≥80% of pending texts is to ensure that the combined searched texts are closer to the relevant needs of the corresponding search personnel and will not deviate too much, so as to lock in specific similar texts, and then display based on these similar texts to achieve better processing results);

[0050] S35. Based on the marked similar texts, identify the number of searches for the corresponding similar texts in the historical data, and sort the search numbers from large to small, select the similar texts associated with the first four groups, and use the similar texts and the determined characteristic texts as synchronous search texts (the so-called combined search text means that each corresponding text is searched for content, ensuring that each search content can be searched within the corresponding synchronous timeline, which is convenient for search personnel to compare and view, etc.);

[0051] Step 4: Based on the determined synchronous search text, execute a multi-threaded synchronous search mode, synchronously search different texts in the synchronous search text, and monitor the search rate in real time. Based on the analysis results of the real-time monitoring, adjust the computing power associated with different threads in real time to ensure synchronous output of the synchronous search text;

[0052] The specific sub-steps for real-time adjustment of the associated computing power of different threads are:

[0053] S41, according to the number of texts in the synchronous search text, the same number of threads are executed, and based on the computing power held by the artificial intelligence, the computing power is evenly divided according to the number of threads, so that each thread can evenly share the same computing power, and each different thread searches for different texts according to the evenly divided computing power, and based on the search process, the search rate is monitored in real time;

[0054] S42, calibrate the retrieval rate monitored in real time by each different thread as SV q , where q represents different threads, and multiple sets of retrieval rates SV monitored in real time q Perform variance processing, confirm the rate variance F, and check the rate variance F with the preset value Y1:

[0055] If F>Y1, the thread associated with the highest search rate is confirmed and marked as the high thread, and the thread associated with the lowest search rate is marked as the low thread. The computing power is allocated from the high thread to the low thread until the associated rate variance F≤Y1 is stopped. If the search rate of the high thread is not the highest value or the search rate of the low thread is not the lowest value, the high thread or the low thread is reconfirmed, and the computing power is allocated based on the reconfirmed high thread and low thread. The high thread and the low thread will change with the specific computing power allocation process. That is, during the allocation process, the high thread will allocate computing power to the low thread, and its high thread will change from high to low with the corresponding computing power allocation. Then the original high thread’s retrieval rate is no longer the highest state, so the high thread and the low thread with the highest search rate are re-determined, and then the allocation continues;

[0056] If F≤Y1, determine the current moment, and based on the current moment, confirm the retrieval rate generated by the corresponding thread in the past moment, and confirm the retrieval rate curve generated by the corresponding thread based on the timeline. From this retrieval rate curve, determine the change value between adjacent moments, and its change value = retrieval rate of the next moment - retrieval rate of the previous moment (the retrieval rates corresponding to adjacent moments 1 and 2 are proposed to be J1 and J2 respectively, where 2 is after 1, then the change value between adjacent moments 1 and 2 = J2-J1), and select the maximum change value and the minimum change value from the confirmed groups of change values, based on the maximum change value, the minimum change value and the retrieval rate SV associated with the current moment q , predict the highest search rate and the lowest search rate of the corresponding thread at the next moment, and the highest search rate = SV q + Maximum change value, minimum retrieval rate = SV q +Minimum change value;

[0057] The highest retrieval rate and the lowest retrieval rate associated with each thread at the next moment are confirmed in the same manner;

[0058] The highest or lowest search rate in each group of threads is randomly combined. When combining, each group of threads selects a group of search rates. The combined groups of search rates are processed for variance to identify whether there is a combination of F>Y1. If so, the computing power allocation method of F>Y1 is directly used to allocate computing power between threads to avoid the situation of F>Y1. If not, continuous monitoring and variance determination can be performed.

[0059] S43, the text content associated with each text search in the synchronous search text is synchronously displayed for review by the corresponding search personnel (when displaying, the search personnel will first see the corresponding text title, that is, the corresponding similar text or feature text. When clicking on the similar text or feature text, the corresponding text content will be directly displayed);

[0060] Specifically, in order to enable the corresponding search content to be output synchronously and compared, a multi-threaded synchronous search method is adopted to perform real-time search on each text. During the search process, in order to ensure that the output time is relatively consistent, it is prioritized to ensure that the search rate of each thread is relatively consistent, thereby completing the corresponding text content synchronous display process, facilitating relevant search personnel to compare the content, and achieving better processing results.

[0061] Second embodiment

[0062] Combination Figure 2 , a combined retrieval optimization system based on artificial intelligence, comprising:

[0063] The text initial processing end receives and pre-processes the text input by the user, first performs format standardization on the text, and then removes symbols from the text that has completed format standardization to obtain keyword text;

[0064] The feature determination end reprocesses the keyword text based on the determined keyword text, checks the keyword text with a preset word library, locks in and removes words, and partially removes words in the keyword text to confirm the feature text;

[0065] The combined search text determination end, based on the confirmed feature text, locks the feature vector associated with the feature text from the corresponding vocabulary vector library, and then identifies other combined search texts with the feature vector from the past historical search data, selects similar texts from the identified groups of combined search texts, and uses the selected similar texts and the feature text as synchronous search texts;

[0066] The retrieval process optimization end executes a multi-threaded synchronous retrieval method based on the determined synchronous retrieval text, synchronously retrieves different texts within the synchronous retrieval text, and monitors its retrieval rate in real time. Based on the analysis results of real-time monitoring, the computing power associated with different threads is adjusted in real time.

[0067] Some of the data in the above formulas are dimensionless and numerically calculated. Meanwhile, the contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0068] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A combined retrieval optimization method based on artificial intelligence, characterized in that: The following steps are involved: Step 1: Receive and pre-process the text input by the user: first standardize the format of the text to confirm the standard text, then remove the symbols from the standard text to obtain the keyword text; Step 2: reprocess the determined keyword text: check the keyword text with the preset word library, lock the excluded words, remove the excluded words in the keyword text, and confirm the feature text; Step 3: Based on the confirmed feature text, lock the feature vector associated with the feature text from the preset vocabulary vector library, and then identify other combined search texts with the feature vector from the past historical search data, select similar texts from the identified several groups of combined search texts, and use the selected similar texts and the feature text as synchronous search texts; Step 4: Based on the determined synchronous search text, perform synchronous search on different texts in the synchronous search text by executing a multi-threaded synchronous search method, and monitor the search rate in real time. Based on the analysis results of the real-time monitoring, adjust the computing power associated with different threads in real time. The specific sub-steps are: S41, according to the number of texts in the synchronous search text, the same number of threads are executed, and based on the computing power held by the artificial intelligence, the computing power is evenly divided according to the number of threads, so that each thread can evenly share the same computing power, and each different thread searches for different texts according to the evenly divided computing power, and based on the search process, the search rate is monitored in real time; S42, calibrating the retrieval rate monitored in real time by each different thread as SV q , where q represents different threads, and multiple sets of retrieval rates SV monitored in real time q Perform variance processing, confirm the rate variance F, and check the rate variance F with the preset value Y1: If F>Y1, the thread associated with the current highest search rate is confirmed and marked as the high thread, and the thread associated with the current lowest search rate is marked as the low thread. The computing power is allocated from the high thread to the low thread until the associated rate variance F≤Y1 is stopped. If the search rate of the high thread is not the highest value or the search rate of the low thread is not the lowest value, the high thread or the low thread is reconfirmed, and the computing power is allocated according to the reconfirmed high thread and low thread, where the high thread and the low thread will change with the specific computing power allocation process; If F≤Y1, determine the current moment, and confirm the search rate generated by the corresponding thread in the past moments based on the current moment, and confirm the search rate curve generated by the corresponding thread based on the timeline. From this search rate curve, determine the change value between adjacent moments, where the change value = the search rate of the next moment - the search rate of the previous moment, and select the maximum change value and the minimum change value from the confirmed groups of change values, and calculate the search rate SV associated with the current moment based on the maximum change value, the minimum change value, and the current moment. q , predict the highest search rate and the lowest search rate of the corresponding thread at the next moment, and the highest search rate = SV q + Maximum change value, minimum retrieval rate = SV q +Minimum change value; The highest retrieval rate and the lowest retrieval rate associated with each thread at the next moment are confirmed in the same manner; The highest retrieval rate or the lowest retrieval rate in each group of threads is randomly combined. When combining, each group of threads selects a group of retrieval rates. The variance of the combined groups of retrieval rates is processed to identify whether there is a combination of F>Y1. If so, the computing power allocation method of F>Y1 is directly used to allocate computing power between threads. If not, continuous monitoring and variance judgment can be performed.

2. The combined retrieval optimization method based on artificial intelligence according to claim 1, characterized in that: In the step 1, the sub-steps of obtaining the keyword text are: S11. Based on the standard format preset in the artificial intelligence, identify whether the format of the input text is consistent with the standard format. If it is consistent, mark the input text as the standard text. If it is not consistent, perform format brush processing on the input text to adjust the format of the input text to the standard format, and mark the adjusted text as the standard text. S12. Based on a preset symbol database, identify the symbols and spaces in the standard text, and remove the identified special symbols and spaces to obtain the keyword text.

3. The combined retrieval optimization method based on artificial intelligence according to claim 1 is characterized in that: In step 2, the specific sub-steps of confirming the feature text are: Based on the confirmed keyword text, the words in the keyword text are checked with the preset word library, and the related words with the check result are marked as the elimination words, and based on the marked elimination words, the elimination words are eliminated from the keyword text, and the processed keyword text is marked as the feature text; The words stored in the preset word library are all preset words.

4. The combined retrieval optimization method based on artificial intelligence according to claim 1, characterized in that: In step 3, the specific sub-steps of using the selected similar text and feature text as synchronous retrieval text are: S31, based on the confirmed feature text and a preset vocabulary vector library, the vocabulary vector library is a preset vector library, which stores feature vectors corresponding to different words, confirming the feature vector associated with the vocabulary in the feature text from the vocabulary vector library, and then sorting the plurality of groups of feature vectors associated with the feature text in a front-to-back manner according to the vocabulary, and confirming the feature vector set; S32, based on the confirmed feature vector set, confirming from past historical search data a combined search text including all feature vectors in the feature vector set, wherein the multiple groups of feature vectors associated in the combined search text include all feature vectors in the feature vector set and other feature vectors; S33, confirm the position of the feature vector set again, identify the feature vector subsequently associated with the corresponding feature vector in the feature vector set, confirm the feature vector that belongs to the same vector as the feature vector in the feature vector set from the confirmed combined search text, mark the confirmed feature vector as the main vector, hide other feature vectors in the combined search text that are not marked as the main vector, identify whether the position relationship between the multiple groups of main vectors is consistent with the position relationship of the feature vectors in the feature vector set, if consistent, mark the combined search text as the pending text, and display the feature vector hidden in the pending text, if not consistent, no calibration is performed; S34, based on the determined pending text, identify the proportion of the total length of its main vector in the total length of the overall feature vector: mark the length of each main vector as L i-k , where i represents different pending texts, k represents different main vectors in the pending text, and the lengths of several main vectors L i-k Perform summation to confirm the first total value ZL i , and then determine the length of all feature vectors in the pending text and perform summation to confirm the second sum value BL i , using ZB i =ZL i ÷BL i Confirm the percentage value ZB associated with this pending text i , will satisfy: ZB i ≥80% of the pending texts are marked as similar texts, otherwise no marking is performed; S35. Based on the marked similar texts, identify the number of searches for the corresponding similar texts in the historical data, and sort the search times from large to small, select the similar texts associated with the first four groups, and use the similar texts and the determined feature texts as synchronous search texts.

5. The combined retrieval optimization method based on artificial intelligence according to claim 1 is characterized in that: The step 4 also includes: S43. The text contents associated with each text search in the synchronous search text are synchronously displayed for review by the corresponding search personnel.

6. A combined search optimization system based on artificial intelligence, which operates based on the combined search optimization method based on artificial intelligence according to any one of claims 1 to 5, characterized in that: include: The text initial processing end receives and pre-processes the text input by the user, first performs format standardization on the text, and then removes symbols from the text that has completed format standardization to obtain keyword text; The feature determination end reprocesses the keyword text based on the determined keyword text, checks the keyword text with a preset word library, locks in and removes words, and partially removes words in the keyword text to confirm the feature text; The combined search text determination end, based on the confirmed feature text, locks the feature vector associated with the feature text from the preset vocabulary vector library, and then identifies other combined search texts with the feature vector from the past historical search data, selects similar texts from the identified groups of combined search texts, and uses the selected similar texts and the feature text as synchronous search texts; The retrieval process optimization end executes a multi-threaded synchronous retrieval method based on the determined synchronous retrieval text, synchronously retrieves different texts within the synchronous retrieval text, and monitors its retrieval rate in real time. Based on the analysis results of real-time monitoring, the computing power associated with different threads is adjusted in real time.

Images