Tumor guide medication information generation system and method, electronic equipment and medium

Abstract

The embodiment of the invention discloses a tumor guide medication information generation system and method, electronic equipment and a medium. The system comprises a data acquisition server used for acquiring a tumor guide medication information set; carrying out classification processing on the obtained tumor guide medicine recommendation query statement information set; the cue word generation server is used for generating a content extraction cue word information set; the medication information analysis server is used for performing information analysis processing on the tumor guide medication information set; the tumor drug use database construction server is used for performing drug knowledge extraction processing on the tumor guide drug use analysis information set to obtain a tumor drug use knowledge base; the tumor medication retrieval server is used for generating a tumor guide recommended medication information set and sending the tumor guide recommended medication information set to the user interaction terminal; and the user interaction terminal is used for terminal display. According to the embodiment, the construction time of the tumor drug knowledge base and the display waiting time of the drug recommendation information can be shortened, the waste of display and transmission resources is reduced, and the load of each server is reduced.

Description

TECHNICAL FIELD

[0001] Embodiments of the present disclosure relate to the technical field of computer technology, and in particular to a tumor guideline medication information generation system, method, electronic device and medium. BACKGROUND

[0002] Tumor diagnosis and treatment guidelines are important evidence-based references for anti-tumor drug treatment. However, tumor guidelines are widely sourced, have various cancer types, complex structures and are updated rapidly. The guideline medication recommendation information content is complex and has characteristics such as multiple drug name references, complex population stratification conditions, and complex traceability paths. For the generation of tumor guideline medication information, the commonly used method is as follows: a tumor medication knowledge base server is used to input tumor guideline medication information and prompt word information generated by interacting with a large language model into the large language model to generate a tumor medication knowledge base. Then, an information retrieval server is used to match keywords in the tumor medication knowledge base through search keywords input by a user interaction terminal to obtain tumor medication recommendation information for a target user. The tumor medication recommendation information is sent to the user interaction terminal. Finally, the user interaction terminal is used to display the tumor medication recommendation information.

[0003] However, in practice, it is found that when the above method is used to generate tumor guideline medication information, the following technical problems often exist:

[0004] The display waiting time of the medication recommendation information is long. After the user inputs the query content through the user interaction terminal, the user input query content does not exist in the currently constructed tumor medication knowledge base, and the prompt word needs to be generated by repeatedly interacting with the large language model and then input into the large language model to construct the tumor medication knowledge base. The hallucination risk of the large language model and the multiple interactions increase the construction time of the tumor medication knowledge base, and further increase the waiting time of the medication recommendation information sent to the front end display;

[0005] The display information of the medication recommendation information is redundant and wastes a large amount of display resources. Since only keyword retrieval matching is performed with the tumor medication knowledge base, the retrieval information considered is relatively single, and the user's query intent cannot be accurately identified, resulting in a large amount of incorrect medication recommendation information in the user's retrieved medication recommendation information, which wastes a large amount of display resources when displayed on the user interaction terminal;

[0006] The load of each server is increased. Since the tumor medication knowledge base server needs to run the large language model multiple times, a large amount of computing resources are required to support the running of the large language model, which increases the load of the server. Since the information retrieval server generates a large amount of incorrect information, it needs to repeatedly communicate with the user interaction terminal to send the medication recommendation information, which increases the load of the information retrieval server and the user interaction terminal.

[0007] The above information disclosed in this Background section is only for enhancing the understanding of the background of the present disclosure, and therefore, it can contain information that does not form the prior art that is already known in this country to those of ordinary skill in the art. SUMMARY

[0008] The summary of the present disclosure is intended to introduce the general concepts of the present disclosure in a simplified form, which will be described in detail in the subsequent detailed description section. The summary of the present disclosure is not intended to identify key or essential features of the claimed technology nor is it intended to be used to limit the scope of the claimed technology.

[0009] Some embodiments of the present disclosure propose a tumor guideline drug information generation system, method, electronic device and medium to solve one or more of the technical problems mentioned in the background section.

[0010] In a first aspect, some embodiments of the present disclosure provide a tumor guideline drug information generation system, comprising a data collection server, a prompt word generation server, a drug information analysis server, a tumor drug database construction server, a tumor drug search server and a user interaction terminal, wherein: the data collection server is configured to obtain a tumor guideline drug information set and a tumor guideline drug recommendation query sentence information set; the tumor guideline drug recommendation query sentence information set is classified to obtain a tumor query sentence grouping information group set; the prompt word generation server is configured to generate a content extraction prompt word information set for the tumor guideline drug information set according to the tumor query sentence grouping information group set; the drug information analysis server is configured to perform information analysis processing on the tumor guideline drug information set according to the content extraction prompt word information set to obtain a tumor guideline drug analysis information set; the tumor drug database construction server is configured to perform drug knowledge extraction processing on the tumor guideline drug analysis information set according to the content extraction prompt word information set and a guideline content extraction large language model to obtain a tumor drug knowledge base; the tumor drug search server is configured to generate a tumor guideline recommended drug information set according to a tumor guideline drug recommendation query sentence information set input by the user interaction terminal and the tumor drug knowledge base; the tumor guideline recommended drug information set is sent to the user interaction terminal; and the user interaction terminal is configured to display the tumor guideline recommended drug information set in the form of voice or text.

[0011] In a second aspect, some embodiments of the present disclosure provide a method for generating tumor guideline drug information, comprising: obtaining a tumor guideline drug information set and a tumor guideline drug recommendation query sentence information set; performing classification processing on the tumor guideline drug recommendation query sentence information set to obtain a tumor query sentence grouping information group set; generating a content extraction prompt word information set for the tumor guideline drug information set according to the tumor query sentence grouping information group set; performing information analysis processing on the tumor guideline drug information set according to the content extraction prompt word information set to obtain a tumor guideline drug analysis information set; performing drug knowledge extraction processing on the tumor guideline drug analysis information set according to the content extraction prompt word information set and a guideline content extraction large language model to obtain a tumor drug knowledge base; generating a tumor guideline recommended drug information set according to the tumor guideline drug recommendation query sentence information set input by a user interactive terminal and the tumor drug knowledge base; sending the tumor guideline recommended drug information set to the user interactive terminal; and displaying the tumor guideline recommended drug information set on the terminal in the form of voice or text.

[0012] In a third aspect, some embodiments of the present disclosure provide an electronic device, comprising: one or more processors; a storage device having one or more programs stored thereon, when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation manner of the first aspect.

[0013] In a fourth aspect, some embodiments of the present disclosure provide a computer readable medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the method described in any implementation manner of the first aspect.

[0014] The above various embodiments of the present disclosure have the following beneficial effects: the tumor guideline drug information generation system of some embodiments of the present disclosure can shorten the tumor drug knowledge base construction time and the drug recommendation information display waiting time, reduce the waste of display and transmission resources, and reduce the load of each server. Specifically, the reasons for the long display waiting time of the related drug recommendation information, the redundancy of the display information of the drug recommendation information, the waste of a large amount of display resources, and the increase of the load of each server are as follows: the long display waiting time of the drug recommendation information. After the user inputs the query content through the user interaction terminal, there is no user input query content in the currently constructed tumor drug knowledge base, and the large language model needs to be repeatedly interacted and evaluated to generate prompt words and then input to the large language model again to construct the tumor drug knowledge base. The large language model has the illusion risk and multiple interactions increase the construction time of the tumor drug knowledge base, which further increases the waiting time of the drug recommendation information sent to the front end display; the display information of the drug recommendation information is redundant, which wastes a large amount of display resources. Since only the keywords are matched with the tumor drug knowledge base, the search information considered is relatively single, which cannot accurately identify the user's query intention, resulting in a large amount of incorrect recommended drug information in the user's search drug recommendation information, which wastes a large amount of display resources when displayed on the user interaction terminal; and increases the load of each server. Since the tumor drug knowledge base server needs to repeatedly run the large language model, a large amount of computing resources are required to support the running of the large language model, which increases the load of the server. Since the information retrieval server generates a large amount of incorrect information, it needs to repeatedly communicate with the user interaction terminal to send recommended drug information, which increases the load of the information retrieval server and the user interaction terminal. Based on this, the tumor guideline drug information generation system of some embodiments of the present disclosure includes a data acquisition server, a prompt word generation server, a drug information analysis server, a tumor drug database construction server, a tumor drug retrieval server, and a user interaction terminal, wherein: the above data acquisition server is used to first acquire a tumor guideline drug information set and a tumor guideline drug recommendation query sentence information set. Here, the tumor guideline drug information set is used for subsequent content extraction, and the tumor guideline drug recommendation query sentence information set is used for subsequent generation of tumor guideline recommended drug information. Secondly, the above tumor guideline drug recommendation query sentence information set is classified to obtain a tumor query sentence grouping information group set. Here, the classification processing can cluster the tumor guideline drug recommendation query sentence information set, which is convenient for subsequent generation of content extraction prompt word information for the guideline content extraction large language model, improves the accuracy and efficiency of the model output. The prompt word generation server is used to generate a content extraction prompt word information set for the above tumor guideline drug information set according to the above tumor query sentence grouping information group set.Here, since the tumor query statement grouping information set can be information with user personalized query style, guiding the large language model to generate expected output content, the accuracy and personalization of the generated content extraction prompt word information set can be improved. The drug information analysis server is configured to perform information analysis processing on the tumor guideline drug information set according to the content extraction prompt word information set, to obtain a tumor guideline drug analysis information set. Here, the logical context and semantic association between different types of layout information included in the different formats of guideline drug information in the tumor guideline drug information set can be comprehensively sorted out, improving the accuracy of the tumor drug guideline analysis information and the understanding ability of the context semantic information. The tumor drug database construction server is configured to perform drug knowledge extraction processing on the tumor guideline drug analysis information set according to the content extraction prompt word information set and the guideline content extraction large language model, to obtain a tumor drug knowledge base. Here, the content extraction prompt word information set can control the output of the guideline content extraction large language model, improve the user experience, reduce the pre-training time of the model, improve the construction efficiency of the tumor drug knowledge base, and at the same time, utilize the powerful context semantic understanding and information integration capability of the guideline content extraction large language model to improve the accuracy, completeness and systematicness of the tumor drug knowledge base, improve the timeliness and efficiency of the tumor drug knowledge base construction, and reduce the construction time. The tumor drug retrieval server is configured to first generate a tumor guideline recommended drug information set according to the tumor guideline drug recommendation query statement information set input by the user interactive terminal and the tumor drug knowledge base. Here, based on the comprehensive and accurate tumor drug knowledge base, the accuracy of the generated tumor guideline recommended drug information set can be improved, the amount of error and redundant tumor guideline recommended drug information data can be reduced, manual screening by the user can be avoided, and the user experience can be improved. Then, the tumor guideline recommended drug information set is sent to the user interactive terminal. Here, the waste of communication resources can be reduced. The user interactive terminal is configured to display the tumor guideline recommended drug information set on the terminal in the form of voice or text, reducing the load of each server. Here, the display waiting time of the recommended drug information can be shortened. Therefore, the tumor guideline drug information generation system can shorten the tumor drug knowledge base construction time and the recommended drug information display waiting time, reduce the waste of display and transmission resources, and reduce the load of each server. BRIEF DESCRIPTION OF DRAWINGS

[0015] The above and other features, advantages, and aspects of embodiments of the present disclosure will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings in which:

[0016] Figure 1is a timing diagram of some embodiments of the tumor guideline medication information generation system according to the present disclosure;

[0017] Figure 2 is a page schematic diagram of searching and outputting guideline recommended medication information according to cancer type and tumor drug name in some embodiments of the tumor guideline medication information generation system according to the present disclosure;

[0018] Figure 3 is a page schematic diagram of different search condition display of patient population classification information under cancer type search condition in some embodiments of the tumor guideline medication information generation system according to the present disclosure;

[0019] Figure 4 is a page schematic diagram of searching for guideline recommended medication information by a user in the form of a sentence in a natural language search mode in some embodiments of the tumor guideline medication information generation system according to the present disclosure;

[0020] Figure 5 is a flowchart of some embodiments of the tumor guideline medication information generation method according to the present disclosure;

[0021] Figure 6 is a structural schematic diagram of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0022] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although some embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms, and should not be interpreted as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly and completely understood. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes, and are not intended to limit the scope of protection of the present disclosure.

[0023] In addition, it should be further noted that only parts related to the present application are shown in the drawings for ease of description. The embodiments in the present disclosure and the features in the embodiments can be combined with each other without conflict.

[0024] It should be noted that the concepts of "first", "second", etc. mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not intended to limit the order or interdependence of the functions performed by these devices, modules or units.

[0025] It should be noted that the modification of "one", "multiple" in the present disclosure is illustrative and not restrictive, and those skilled in the art should understand that unless otherwise explicitly indicated in the context, it should be understood as "one or more".

[0026] Names of messages or information exchanged between multiple devices in embodiments of the present disclosure are only for illustrative purposes, and are not intended to limit the scope of the messages or information.

[0027] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0028] Figure 1 A timing diagram showing some embodiments of a tumor guideline drug information generation system according to the present disclosure is shown.

[0029] As Figure 1 shown, a tumor guideline drug information generation system includes a data collection server 101, a prompt word generation server 102, a drug information analysis server 103, a tumor drug database construction server 104, a tumor drug search server 105, and a user interaction terminal 106. Among them, the data collection server and the prompt word generation server are communicatively connected. The prompt word generation server and the tumor drug database construction server are communicatively connected. The drug information analysis server and the prompt word generation server are communicatively connected. The tumor drug database construction server and the tumor drug search server, the data collection server are communicatively connected. The user interaction terminal and the tumor drug search server are communicatively connected.

[0030] Among them, the data collection server 101 is configured to obtain a tumor guideline drug information set and a tumor guideline drug recommendation query sentence information set.

[0031] In some embodiments, the tumor guideline drug information generation system described above can obtain the tumor guideline drug information set and the tumor guideline drug recommendation query sentence information set through a wired connection or a wireless connection. Among them, the tumor guideline drug information in the tumor guideline drug information set can be text information providing different drug treatment plans for different types of tumors. For example, the tumor guideline drug information set can include but is not limited to at least one of the following: CSCO (Chinese Society of Clinical Oncology, Chinese Society of Clinical Oncology) guideline recommended drug information, NCCN (National Comprehensive Cancer Network, American National Comprehensive Cancer Network) guideline recommended drug information. The tumor guideline drug recommendation query sentence information in the tumor guideline drug recommendation query sentence information set can be sentence information collected by the user interaction terminal for querying the guideline recommended drug treatment information included in the tumor guideline recommended drug information.

[0032] The data collection server 101 is configured to classify the tumor guideline drug recommendation query sentence information set to obtain a tumor query sentence grouping information group set.

[0033] In some embodiments, the execution subject can classify the tumor guideline drug recommendation query sentence information set to obtain a tumor query sentence grouping information group set. The tumor query sentence grouping information in the tumor query sentence grouping information group set can be query sentence information grouped according to different search conditions. For example, the tumor query sentence grouping information group set can include, but is not limited to, at least one of the following: grouping information obtained by grouping according to keywords, grouping information obtained by grouping according to cancer query text sentence forms. In practice, the execution subject can first classify the tumor guideline drug recommendation query sentence information set according to the number of tumor guideline drugs included in the tumor guideline drug recommendation query sentence information to obtain a first tumor query sentence grouping information group set. The number of tumor guideline drugs can be one or more. Then, for each first tumor query sentence grouping information group in the first tumor query sentence grouping information group set, the first tumor query sentence grouping information group is classified according to the query form of the first tumor query sentence grouping information to obtain a tumor query sentence grouping information group set. The query form classification can include, but is not limited to, at least one of the following: a query form with a predetermined number of options for a multiple-choice question, a query form with specific example context semantic information, and a query form with keywords. The predetermined number can be determined according to specific circumstances and is not limited herein. The specific example can be a query example of the same type as the query sentence for illustration.

[0034] The prompt word generation server 102 is configured to generate content extraction prompt word information set for the tumor guideline drug information set according to the tumor query sentence grouping information group set.

[0035] In some embodiments, the execution subject can group the tumor query sentence grouping information sets according to the tumor query sentence grouping information sets to generate a content extraction prompt word information set for the tumor guideline drug information set. The content extraction prompt word information in the content extraction prompt word information set can be text information used to guide knowledge extraction of the tumor guideline drug information set and output in a constraint format. The content extraction prompt word information can be prompt word information designed according to different categories of tumor query sentence information. For example, the content extraction prompt word information set can include at least one of the following: prompt word information for querying drug information for single and multiple tumor drugs, and prompt word information for querying drug information with specific example context semantic information. For example, the prompt word information for querying drug information with specific example context semantic information can be "known text with the following drug treatment plan: {original material text}, please determine whether {drug name} is part of the drug plan, if yes, answer yes, otherwise answer no. Specific examples are as follows: {example 1}; {example 2}...". As an example, the execution subject can first extract task target information and output format information from the tumor query sentence grouping information set to obtain query task target information and query output format information. Using ICL (In-Context Learning) technology, the content extraction prompt word information set for the tumor guideline drug information set and the multi-modal tumor association information set is generated according to the query task target information and the query output format information.

[0036] In some optional implementations of some embodiments, the prompt word generation server is configured to:

[0037] First, determine the thought chain query decomposition step set of the tumor query sentence grouping information set. The thought chain query decomposition step in the thought chain query decomposition step set can be a step that guides the guideline content extraction large language model to step-by-step decompose the problem corresponding to the tumor query sentence grouping information. The determination can be made using CoT (Chain-of-Thought) technology.

[0038] Second, determine the prompt word category information of each tumor query sentence grouping information set in the tumor query sentence grouping information set. The prompt word category information can represent the query category form information to which the tumor query sentence grouping information set belongs. The prompt word category information can include at least one of the following: single tumor drug query category information, multiple tumor drug query category information, and keyword query category information.

[0039] In the third step, the prompt word category information, the tumor guideline drug information set, and the thought chain query decomposition step group set are input into a prompt word generation model to obtain a content extraction prompt word information set. The prompt word generation model can be a generative large language model that performs semantic association on the input prompt word category information, user emotion information group set, tumor guideline drug information set, and thought chain query decomposition step to generate content extraction prompt words. For example, the prompt word generation model can be a GPT (Generative Pre-trained Transformer) model.

[0040] Further, in the process of using the technical solutions to solve the technical problems mentioned in the background, the following technical problems often occur: the prompt words used in the guide large language model become more and more complex and redundant, causing the attention of the model to be more dispersed, the analysis time to be longer, and the accuracy of the model output information to be lower, which wastes a large amount of server computing resources and increases the load of the server. To solve the above technical problems, the conventional solution is generally to use a large language model to repeatedly evaluate and optimize the initial prompt words output by the user to obtain optimized prompt word information. However, the conventional solution still has the following problems: increasing the generation time of the prompt words and increasing the load of the prompt word server. Because of the need for repeated interaction with the large language model and the need to repeatedly run the large language model, the load of the prompt word generation server is increased and a large amount of memory resources are used. In addition, the large language model has the problem of model hallucination, which reduces the accuracy of the generated prompt words, increases the generation time of the prompt words, and increases the load of the prompt word server. Considering the shortcomings of the conventional solution, the following solution is adopted:

[0041] In some optional implementations of some embodiments, the tumor drug database construction server is configured to:

[0042] In the first step, for each content extraction prompt word information in the content extraction prompt word information set, the following compression optimization steps are performed:

[0043] Sub-step 1, according to the obtained user historical tumor prompt word information set and the tumor field prompt word information set, the content extraction prompt word information is updated to obtain the tumor prompt word information. Among them, the user historical tumor prompt word information in the user historical tumor prompt word information set can be the text information generated by the user before the current time to guide the construction of the knowledge base of the guide content extraction large language model. The tumor field prompt word information in the tumor field prompt word information set can be the prompt word information about the tumor field generated by the expert experience of the tumor expert. In practice, the execution subject can utilize the bidirectional maximum matching algorithm to accurately match the tumor field prompt word information set and the content extraction prompt word information, and utilize the Word2Vec model to perform fuzzy matching to obtain a matching result set. Then, for at least one matching result in the matching result set representing a successful match, the content extraction prompt word set corresponding to the at least one matching result is replaced with tumor terminology by utilizing the tumor terminology mapping table, and the polysemous words in the content extraction prompt word information are subjected to context constraint and ambiguity resolution to obtain the tumor field prompt word information. Among them, the tumor terminology mapping table can be a pre-constructed mapping data table of tumor field terminology-user common expression. Finally, the user historical tumor prompt word information set and the tumor field prompt word information set are subjected to user intent and word habit extraction by utilizing the BERT model to obtain user intent habit information and field intent habit information. Finally, the tumor field prompt word information is modified by utilizing the user intent habit information and the field intent habit information to obtain the tumor prompt word information.

[0044] Sub-step 2, input the prompt word token set corresponding to the tumor prompt word information into the context semantic extraction model to obtain a prompt word semantic feature vector set. Among them, the prompt word semantic feature vector in the prompt word semantic feature vector set can represent the context semantic information of the prompt word in the tumor prompt word information. The context semantic extraction model can be a model for extracting the semantic of the input prompt word token. The context semantic extraction model can be a Transformer model. The prompt word token in the prompt word token set can be a word obtained by tokenizing the content extraction prompt word information.

[0045] In sub-step 3, the prompt token perplexity of each prompt semantic feature vector in the set of prompt semantic feature vectors is determined to obtain a set of prompt token perplexities. The prompt token perplexity can represent the importance of the prompt semantic feature vector in the content extraction prompt information. In practice, the subject can first determine the appearance probability of the prompt token segmentation corresponding to each prompt semantic feature vector in the set of prompt semantic feature vectors in the tumor prompt information, to obtain a set of appearance probabilities. The appearance probability can be the ratio of the number of times of the prompt token segmentation appearing and the sum of the number of times of each prompt token segmentation appearing. Then, the reciprocal of the average of the cumulative sum of the logarithmic function values of each appearance probability in the set of appearance probabilities with 10 as the base is determined as the segmentation index value. The cumulative sum is accumulated for the same prompt token segmentation. Finally, the exponential function value with the segmentation index value as the index and 2 as the base is determined as the prompt token perplexity.

[0046] Sub-step 4, according to the above prompt token perplexity set, determine the text correlation degree value set of the tumor guideline drug text segment information set and the tumor guideline drug recommendation query sentence information set corresponding to the above tumor prompt information. Wherein, the tumor guideline drug text segment information in the above tumor guideline drug text segment information set can be the text segment information obtained by text clustering and dividing the above tumor guideline drug information. The text correlation degree value in the above text correlation degree value set can represent the correlation degree of the tumor guideline drug information and the tumor guideline drug recommendation query sentence information. The larger the text correlation degree value is, the more relevant the tumor guideline drug information and the tumor guideline drug recommendation query sentence information are. In practice, the execution subject can execute the following steps for each tumor guideline drug text segment information: first, remove at least one prompt token from the above tumor prompt information whose corresponding prompt token perplexity is less than a preset token perplexity threshold to obtain target extraction prompt information. Wherein, the above preset token perplexity threshold can be a critical value preset to determine whether the prompt token is important in the tumor prompt information. It should be noted that the above preset token perplexity threshold needs to be determined according to the specific situation, which is not limited here. Then, constraint information extraction is performed on the tumor guideline drug recommendation query sentence information set corresponding to the above target extraction prompt information and the above target extraction prompt information to obtain a constraint text information set. Wherein, the constraint text in the above constraint text information set can be information used to constrain the output of the model. For example, the above constraint text information can be "the answer must come from the given document". Then, determine the constraint token condition probability set of each constraint token in the above tumor guideline drug text segment information set corresponding to each constraint text information in the above constraint text information set. Finally, determine the inverse number of the mean of the cumulative sum of the logarithmic function values of each constraint token condition probability in the above constraint token condition probability set as a true number with base 10 as the text initial compression ratio set of the above tumor guideline drug text segment information set.

[0047] Sub-step 5, determine the text initial compression ratio set of the above tumor guideline drug text segment information set through the above text correlation degree value set. Wherein, the text initial compression ratio in the above text initial compression ratio set can represent the compression degree of the internal redundant text of the tumor guideline drug information. In practice, the execution subject can first reorder the above tumor guideline drug text segment information set according to the order of the above text correlation degree value set from large to small to obtain a tumor guideline drug text segment information sequence. Then, determine the text initial compression ratio of the above tumor guideline drug text segment information sequence through artificial experience value.

[0048] Sub-step 6, according to the text initial compression ratio set and the text correlation value sequence corresponding to the tumor guideline medication text segment information sequence, determine the dynamic compression ratio sequence of the tumor guideline medication text segment information sequence. Wherein, the dynamic compression ratio can represent the information compression degree of the tumor guideline medication text segment information. As an example, the execution subject can execute the following determination steps for each tumor guideline medication text segment information in the tumor guideline medication text segment information sequence: first, determine the product of 2 and the position index of the tumor guideline medication text segment information in the tumor guideline medication text segment information sequence as a first value. Secondly, determine the ratio of the first value and the number of tumor guideline medication text segment information included in the tumor guideline medication text segment information sequence as a second value. Then, determine the product of the difference between 1 and the second value and the preset compression control factor as a third value. Wherein, the preset compression control factor can represent a preset control step of dynamically allocated compression ratio. Then, determine the sum of the third value and the corresponding text initial compression ratio as a fourth value. After that, select the minimum value from the fourth value and 1 to determine the fifth value. Finally, select the maximum value from the fifth value and 0 to determine the dynamic compression ratio.

[0049] Sub-step 7, through the dynamic compression ratio sequence, the tumor prompt word information is compressed to obtain the compressed extracted prompt word information as the optimized content prompt word information, and according to the optimized content prompt word information set and the guideline content extraction large language model, the tumor medication guideline analysis information set is subjected to drug knowledge extraction processing to obtain the tumor medication knowledge base.

[0050] The technical solution and related content thereof serve as one invention point of the embodiments of the present disclosure, and solve the technical problem of increasing prompt word generation time, wasting server computing resources, and increasing server load. The factors that cause the increase in prompt word generation time, waste of server computing resources, and increase in server load are often as follows: in the process of guiding the large language model for carding, the prompt word changes are increasingly complex and redundant, causing the model's attention to be more dispersed, the analysis time to be increasingly long, and the large language model to need to be repeatedly run, which increases the load of the prompt word generation server and uses a large amount of memory resources, resulting in lower accuracy of model output information, waste of server computing resources, and increase in server load. If the above factors are solved, the effects of shortening the prompt word generation time, reducing the waste of server computing resources, and reducing the server load can be achieved. To achieve this effect, the present disclosure performs the following compression optimization steps on each content extraction prompt word information in the content extraction prompt word information set: first, the user history tumor prompt word information set and the tumor field prompt word information set are updated to obtain more user habit-compliant and tumor field professional prompt word information, which can improve the quality of the prompt words and the user habit-compliant words, improve the user personalization and professionalization of the prompt words. Second, the content extraction prompt word information is segmented and input into the context semantic extraction model to determine the prompt word token perplexity, which can determine the importance of each segmented word by analyzing the context information of the segmented prompt words. Third, the tumor prompt word information is segmented and removed by the prompt word token perplexity to initially screen out redundant information, which can reduce the data volume and improve the subsequent compression efficiency. Subsequently, the text relevance value set corresponding to the tumor guideline medication text segment information set and the target extraction prompt word information is determined, which can retain the tumor guideline medication text segments that are highly relevant to the tumor guideline medication recommendation query sentence information, improve the relevance of the tumor prompt word information, and further reduce the redundant information of the tumor prompt word information. Subsequently, the tumor guideline medication text segment information set is reordered, the tumor guideline medication text segments with higher importance are placed in front of the prompt word information, the position bias of model analysis and processing is reduced, the perception of key information is improved, and key information can be effectively avoided.Finally, the initial text compression ratio set is determined, and the dynamic compression ratio sequence of the tumor guideline medication text segment information sequence is determined through the text correlation degree value sequence, the optimized content prompt word information set is obtained, which can improve the overall compression efficiency of the tumor guideline medication text segment information, can improve the degree of simplification of the optimized content prompt word information and reduce the amount of prompt word data on the basis of ensuring accuracy, and according to the optimized content prompt word information set and the guideline content extraction large language model, the drug knowledge extraction processing is performed on the tumor medication guideline analysis information set to obtain the tumor medication knowledge base. Due to the optimized content prompt word information set, unnecessary calculation of content extraction prompt word information input into the guideline content extraction large language model can be reduced, the load of the server can be reduced, and thus the construction time of the tumor medication knowledge base can be shortened.

[0051] The medication information analysis server 103 is configured to perform information analysis processing on the tumor guideline medication information set according to the content extraction prompt word information set to obtain a tumor guideline medication analysis information set.

[0052] In some embodiments, the execution subject can perform information analysis processing on the tumor guideline medication information set according to the content extraction prompt word information set to obtain a tumor guideline medication analysis information set. The tumor guideline medication analysis information in the tumor guideline medication analysis information set can represent guideline recommended drug treatment information obtained by performing semantic correlation relationship on each component part included in the guideline page set of the tumor guideline medication information set.

[0053] In some optional implementations of some embodiments, the medication information analysis server is configured to:

[0054] In the first step, the tumor guideline drug information set is processed in blocks to obtain a tumor drug text information set, a tumor drug table set and a tumor drug flowchart set. The tumor drug text information in the tumor drug text information set can be text information that displays tumor recommended drug information in the form of text. The tumor drug table in the tumor drug table set can be table information that displays tumor recommended drug information in the form of a table. The tumor drug flowchart in the tumor drug flowchart set can be chart information that displays the process of tumor recommended drug information in the form of a flowchart. As an example, the execution subject can first convert the format of the tumor guideline drug information set using a PaddleOCR model to obtain a converted tumor guideline drug information set. The converted tumor guideline drug information can be tumor guideline drug information displayed in the form of word. Then, the converted tumor guideline drug information set is processed for text repair to obtain a repaired tumor guideline drug information set. The repaired tumor guideline drug information can be tumor guideline drug information obtained by correcting text characters and repairing cell delimiters. The repair of the cell delimiters can be a delimiter repair using a python-docx package according to a repair strategy of whether the content in the cell is semantically coherent. The repair strategy can be a strategy of merging into one row if the content in the adjacent cells is semantically coherent, and keeping separate rows if the content in the adjacent cells is not semantically coherent. Finally, the repaired tumor guideline drug information set is processed for region detection using a layout analysis technique to obtain the tumor drug text information set, the tumor drug table set and the tumor drug flowchart set.

[0055] In the second step, the tumor drug text information set is processed for character recognition to obtain a tumor drug text character information set. The tumor drug text character information in the tumor drug text character information set can be information that converts tumor drug text information into a character sequence that can be understood and processed by a computer. The character recognition can be text recognition using a RARE (Robust text recognizer with Automatic REctification) model.

[0056] In the third step, according to the above content, the table analysis prompt word information set is extracted, the table text recognition is performed on the tumor drug table set to obtain a tumor drug table information set. The tumor drug table information in the tumor drug table information set can be text information obtained after recognizing the text in the tumor drug table. The table analysis prompt word information can be text information used to guide the large language model to perform table analysis. The table text recognition can include, but is not limited to, at least one of the following: table title recognition, table annotation content recognition, and table content recognition. It should be noted that in the table title recognition, after the large language model receives the input data of the tumor drug table set, the table analysis prompt word information set is used to accurately focus on the features of the text around the table. The large language model quickly extracts the title of the table through image recognition and text analysis capabilities, which can effectively improve the accuracy and efficiency of table title recognition. In the table annotation content recognition, the large language model can accurately recognize the annotation content behind the table under the guidance of the table analysis prompt word information. Then, the large language model further extracts the annotation label in the footnote to accurately extract the detailed information in the annotation and accurately associate and match the cell content in the table. In the case where the annotation text involves references, the reference mark symbol and the reference content are matched, and the paper full text is downloaded by using the crawler technology and the API (Application Programming Interface, application programming interface) interface of the paper database. Then, the downloaded PDF (Portable Document Format, portable document format) document is deeply analyzed by using the large language model to ensure that comprehensive and accurate reference information is obtained, which provides more abundant knowledge support for guideline information analysis.

[0057] In the fourth step, the prompt information set is parsed according to the above content, the tumor drug flowchart set is subjected to flowchart text recognition, and the tumor drug flow information set is obtained. The flowchart parsing prompt information can be text information used to guide the large language model to recognize the flowchart text. The tumor drug flow information can be information representing the tumor drug flowchart in the form of text. It should be noted that the large language model can recognize the elements (such as arrows, icons, and text boxes) in the flowchart according to the flowchart parsing prompt information and the large language model, and describe them in the form of text, which can fully utilize the large language model to parse the flowchart in the tumor guideline drug information set. For arrow elements, the large language model can accurately determine the direction, start and end points, and line characteristics, and further determine the connection and pointing relationship between different elements. For icon and text box elements, the large language model can efficiently extract the text information therein and deeply understand the semantic connotation in combination with the image position relationship. When the flowchart parsing prompt information is “analyze the sequence and logical relationship of each step in the flowchart, and convert the information in the corresponding flowchart into text narrative form”, the large language model can sort out the sequence of the flowchart according to the direction of the arrow, realize efficient utilization and deep mining of the schematic diagram and flowchart information, and further improve the comprehensiveness and effectiveness of the tumor guideline drug information parsing. As an example, the execution subject can output the flowchart parsing prompt information set and the tumor drug flowchart set to the GLM-4V model for flowchart text recognition to obtain the tumor drug flow information set.

[0058] In the fifth step, the tumor drug text information set, the tumor drug table information set, and the tumor drug flow information set are subjected to text correction processing to obtain the corrected tumor drug text information set, the corrected tumor drug table information set, and the corrected tumor drug flow information set. In practice, the execution subject can use a hybrid correction method of statistical language model and regular expression to correct the tumor drug text information set, the tumor drug table information set, and the tumor drug flow information set to obtain the corrected tumor drug text information set, the corrected tumor drug table information set, and the corrected tumor drug flow information set. The statistical language model can be an N-gram model.

[0059] In the sixth step, the corrected tumor drug text information set, the corrected tumor drug table information set, and the corrected tumor drug flow information set are determined as the tumor guideline drug analysis information set.

[0060] Further, in the process of adopting the technical solutions to solve the technical problems mentioned in the background, the following technical problems often accompany: the tumor guideline medication information covers a large amount of population stratification condition information distributed in multiple tables, such as pathological type, disease stage, treatment line, and gene type, and the population stratification condition information is widely distributed, and the tables have problems such as complex traceability path, multi-layer nested tables, cross-page content association, and cross-reference of different population treatment plans, resulting in low content accuracy in the table content information extraction process, a large amount of position association errors or extracted information errors in the table data, and an increase in the construction time of the tumor medication knowledge base and a waste of a large amount of database storage resources. In view of the above technical problems, the conventional solution is generally: through cTAKES (Clinical Text Analysis and Knowledge Extraction System), text recognition is performed on the table content included in the tumor guideline medication information to obtain a tumor medication table information set. Then, the tumor medication table information set is input into a database to construct a tumor medication knowledge base. However, the above conventional solution still has the following problems: since cTAKES cannot recognize the context position of the entity after recognizing the tumor entity, and cannot solve the accurate recognition of the table text information of the cross-page content association and the multi-layer nested table, there are a large amount of redundant table text information, resulting in poor quality of the constructed tumor medication knowledge base, prolonging the construction time of the tumor medication knowledge base, and wasting a large amount of storage resources. Considering the shortcomings of the above conventional solution, the following solution is adopted:

[0061] In some optional implementations of some embodiments, the above medication information analysis server is configured to:

[0062] First, according to the table analysis prompt word information set, tumor drug entity information is recognized from the tumor medication table set to obtain a tumor drug entity information set. The tumor drug entity information in the tumor drug entity information set can be the disease state information of the tumor cancer type suffered by the patient. The tumor drug entity information set can include but is not limited to at least one of the following: pathological type, disease stage, treatment line, gene type, PS (Performance Status) score, and past treatment. As an example, the execution subject can input the table analysis prompt word information set and the tumor medication table set into a large language model to obtain the tumor drug entity information set.

[0063] Secondly, the table text block semantic information set is obtained by performing context semantic analysis on the table text sentence information corresponding to the tumor drug entity information set. The table text sentence information can be the text sentence information in the table where the tumor drug entity information is located. The table text block semantic information in the table text block semantic information set can represent the entity association relationship between the tumor drug entity information. The entity association relationship can include at least one of the following: conditional relationship, parallel relationship. In practice, the execution subject can first perform dependency syntax analysis on the table text sentence information to obtain a sentence dependency syntax tree. Then, the entity semantic role information of each node in the sentence dependency syntax tree is determined by using a T5 (Text-To-Text Transfer Transformer) medical semantic role labeling model. The T5 medical semantic role labeling model can be a deep neural network that identifies the semantic role of each tumor drug entity information included in the input table text sentence information. The T5 medical semantic role standard model can be a model including a concatenated input layer, a Transformer encoder, a Transformer decoder, and an output layer. The input layer can be a neural network layer that performs word vector representation after tokenizing the input table text sentence information, while incorporating position encoding. The output layer can be a neural network layer including a Softmax activation function to output probability values of sentence role labels. The sentence role can include at least one of the following: condition, subject, exclusion item, treatment method. Finally, the entity semantic role information set is input into a fully connected layer to classify the association relationship, and the table text block semantic information set is obtained.

[0064] Thirdly, the context position association information set of the tumor drug entity information set in the table text block semantic information set is determined. The context position association information in the context position association information set can represent the position information of the tumor drug entity information in the table text block semantic information.

[0065] In the fourth step, the tumor medication table set is input into a table multi-layer nested text mapping sub-model included in the cross-page table recognition model to obtain a table nested text mapping feature vector set. The cross-page table recognition model further includes a table cross-page text mapping sub-model and a table text cross-reference mapping sub-model. The cross-page table recognition model can be a deep neural network for multi-layer nested recognition, cross-page association recognition, and table text content cross-reference. The table multi-layer nested text mapping sub-model can be a deep neural network model that uses the TableTransformer model to detect table borders and cell coordinates, determines table nesting relationships by coordinate ranges, assigns unique codes to tables to represent parent-child nesting relationships, associates table headers and corresponding columns in parent tables, and generates a "parent table column-child table column-text content" mapping relationship.

[0066] In the fifth step, the table nested text mapping feature vector set is input into the table cross-page text mapping sub-model to obtain a table cross-page text mapping feature vector set. The table cross-page text mapping feature vector in the table cross-page text mapping feature vector set can represent the semantic information of the associated tumor medication tables across pages. In practice, the table cross-page text mapping sub-model can be implemented by the following steps: first, using the Sentence-BERT (Sentence Embeddings using Siamese BERT-Networks) model, the semantic information of the tumor medication table set included in each page of the tumor guide medication information is extracted, including the title, entity, and abstract sentence, to obtain a page table semantic feature vector set. Then, determine the feature cosine similarity values of any two page table semantic feature vectors in the page table semantic feature vector set. Next, determine the number of co-occurring entity information for each feature cosine similarity value corresponding to the page that is greater than or equal to the preset page feature cosine similarity threshold value, to obtain a co-occurring entity number set. Finally, the tumor medication table set of the two pages corresponding to each co-occurring entity number that is greater than or equal to the preset co-occurring number threshold value is selected from the co-occurring entity number set, and the associated text semantic feature is extracted as a table cross-page text mapping feature vector set.

[0067] In the sixth step, the table text cross-reference mapping sub-model is inputted with the table text mapping feature vector set to obtain a tumor drug table information set, and the tumor drug table information set is inputted into a database to construct a tumor drug knowledge base. The table text cross-reference mapping sub-model can first use a regular expression to identify text information with explicit reference relationships in the table, and use a sequence labeling model to identify ambiguous references (for example, the scheme mentioned in the foregoing). Then, for explicit references, the corresponding table number is directly mapped, and for ambiguous references, the tumor drug entity information of the sentence where the reference is located is extracted, the previous content containing the tumor drug entity information is searched, and the reference target is determined through semantic similarity. Finally, the text information with explicit reference relationships is taken as the starting point, the upstream content referenced by the text information and the downstream content referenced by the text information are recursively searched, a chain structure is generated, and a model for obtaining the tumor drug table information set through text recognition is obtained.

[0068] As an inventive point of the embodiment of the present disclosure, the above technical solution and related content solve the technical problem that the tumor guideline medication information covers a large amount of population stratification condition information such as pathological types, disease stages, treatment lines, and gene types distributed in multiple tables, and the population stratification condition information is widely distributed, and the tables have problems such as complex traceability paths, multi-layer nested tables, cross-page content association, and cross-referenced treatment plans for different populations, which leads to low accuracy of the content provided in the table content information extraction process, a large amount of position association errors or extracted information errors in the table data, and an increase in the tumor medication knowledge base construction time and a waste of a large amount of database storage resources. The factors that prolong the construction time of the tumor medication knowledge base and waste a large amount of storage resources are often as follows: the tumor guideline medication information covers a large amount of population stratification condition information such as pathological types, disease stages, treatment lines, and gene types distributed in multiple tables, and the population stratification condition information is widely distributed, and the tables have problems such as complex traceability paths, multi-layer nested tables, cross-page content association, and cross-referenced treatment plans for different populations, which leads to low accuracy of the content provided in the table content information extraction process, a large amount of position association errors or extracted information errors in the table data, and an increase in the tumor medication knowledge base construction time and a waste of a large amount of database storage resources. If the above factors are solved, the construction time of the tumor medication knowledge base can be shortened, and the waste of storage resources can be reduced. In order to achieve this effect, the present disclosure first performs tumor drug entity information recognition through the table analysis prompt word information set and the large language model. The large language model can adaptively learn the tumor drug entity information, can improve the fineness and accuracy of entity recognition, and can avoid the generalization problem of the general knowledge extraction model in the general field. Secondly, the context semantic analysis processing of the table text sentence information is performed, and the context position association information set of the entity is determined, which can analyze the logical relationship (for example, condition limitation, exclusion relationship) of the complex sentence, extract the stratification conditions hidden in the brackets in the nested structure, solve the problem of incomplete condition extraction in the complex sentence, and locate the distribution of the entity in different text positions (title, table, and annotation), and solve the problem that the entity and the context are cut off in the prior art cTAKES and cannot be integrated. Finally, through the cross-page table recognition model, the problems of complex traceability paths, multi-layer nested tables, cross-page content association, and cross-referenced treatment plans for different populations in the tables can be solved, the accuracy and quality of the tumor medication table information set can be improved, the quality of the constructed tumor medication knowledge base can be improved, the waste of storage resources of the tumor medication knowledge base can be reduced, and the construction time of the tumor medication knowledge base can be shortened.

[0069] The tumor medication database construction server 104 is configured to perform drug knowledge extraction processing on the tumor guideline medication analysis information set according to the content extraction prompt word information set and the guideline content extraction large language model, and obtain a tumor medication knowledge base.

[0070] In some embodiments, the execution subject described above can extract the content extraction large language model according to the content extraction prompt word information set and the guideline content, perform drug knowledge extraction processing on the tumor guideline drug analysis information set, and obtain a tumor drug knowledge base. The guideline content extraction large language model can be a large language model that extracts drug information from the input content extraction prompt word information set and the tumor guideline drug information set, and outputs structured information. The large language model can be a large language model in which the self-attention mechanism in the existing large language model (such as GPT, PaLM, LLaMA) is changed to grouped query attention, and a pluggable MoE (Mixture of Experts), a dynamic parameter activation module, and a retrieval enhancement module are added. The grouped query attention can be a grouped query attention that optimizes multi-head attention and reduces memory usage by sharing part of the key-value head. The pluggable MoE can be an expert module that uses dynamic routing to input to different sub-networks to improve model capacity and efficiency. The dynamic parameter activation module can be a parameter activation module that only activates task-related parameters to reduce inference cost. For example, the dynamic parameter activation module can be a Pathways architecture module. The tumor drug knowledge base is a database that stores drug recommendation information for each drug in the form of a data table. The tumor drug knowledge base can include pathological type, tumor stage, treatment line, gene type, PS score, previous treatment, recommended drug regimen, recommended level, and evidence category.

[0071] Optionally, the tumor drug database construction server is configured to:

[0072] In a first step, the tumor drug retrieval term set is determined according to the corrected tumor drug text information set, the corrected tumor drug table information set and the corrected tumor drug process information set included in the above-mentioned tumor guideline drug analysis information set. The tumor drug retrieval term includes at least one of the following: drug generic name, drug alias, drug name abbreviation, drug category, drug regimen name. The tumor drug retrieval term in the tumor drug retrieval term set can be a word representing different drug names of the same tumor recommended drug information. As an example, the execution subject can first determine the drug generic name of the tumor drug by using the existing tumor drug knowledge base. Then, search in the tumor guideline drug analysis information set by drug generic name to obtain the tumor drug retrieval term set. The search can include at least one of the following: table traversal search, cell content search, treatment regimen main component search. The table traversal search can be a traversal search using the python-docx package to traverse the corrected tumor drug table information and locate the column with the recommended level of the guideline recommendation level. The treatment regimen main component search can be used to search whether the appearing drug name is the main content of the recommended drug regimen.

[0073] In a second step, the tumor drug retrieval term set is subjected to semantic fuzzy disambiguation processing to obtain a disambiguated tumor drug retrieval term set. In practice, the execution subject can first perform text table structure analysis on the tumor guideline drug information set by using the PaddleOCR model to identify the content of the table superscript and paragraph to obtain structured text information. Second, the NER (Named Entity Recognition) model is used to perform entity recognition mapping on the tumor drug retrieval term set in the structured text information to obtain a retrieval term vector set, wherein the retrieval term vector can represent the position information and context information of the tumor drug retrieval term. Then, the semantic similarity value group set of the retrieval term vector set and the drug data included in the preset tumor drug database is determined. The preset tumor drug database can be an existing database related to tumor drug information. Finally, at least one tumor drug retrieval term corresponding to at least one semantic similarity value group set with a semantic similarity value greater than or equal to a preset semantic similarity threshold value is determined as the same tumor drug retrieval term. The above-mentioned preset semantic similarity threshold value can be a preset numerical value for determining whether it is the same semantic. For example, the preset semantic similarity threshold value can be 0.85.

[0074] In the third step, the prompt word optimization processing is performed on the extracted prompt word information set to obtain an optimized content prompt word information set. The prompt word optimization processing can include, but is not limited to, at least one of the following: adding a modified case for context optimization, adding a specific type of instruction text, and adding a corresponding rule engine. For example, the specific type of instruction text can be “please remove the related content of the treatment branch {extracted treatment branch}”. The rule engine can be a rule engine for processing low-frequency input (e.g., multiple treatment branches in the text) using regular expressions for pattern recognition and manual processing. It should be noted that the prompt word optimization can further improve the quality and efficiency of drug knowledge extraction.

[0075] In the fourth step, the guideline content extraction large language model is used to extract the drug recommendation information triple set from the corrected tumor drug text information set, the corrected tumor drug table information set, and the corrected tumor drug process information set according to the optimized content prompt word information set and the disambiguated tumor drug search term set in the form of a multiple-choice question. The drug recommendation information triple in the drug recommendation information triple set can be a triple composed of drug recommendation information, field information corresponding to the drug recommendation information, and specific values of the field information. The field information corresponding to the drug recommendation information can include, but is not limited to, at least one of the following: pathological type and tumor stage. It should be noted that the drug attribute extraction can be converted into a task conversion in the form of a multiple-choice question. When extracting the tumor stage to which the tumor drug is applicable, the tumor guideline drug information set is input into the guideline content extraction large language model to extract the tumor stage. For example, the tumor stage can be stage I, stage II, stage III, or stage IV. The multiple-choice question form can effectively narrow the positioning range of the large language model content extraction and assist in improving the accuracy and reliability of the extracted data. As an example, the execution subject can first determine the field information option mapping table included in the disambiguated tumor drug search term set to obtain a tumor drug field option mapping table set. Then, the tumor drug field option mapping table set is added to the optimized content prompt word information set in the form of a multiple-choice question to obtain a selected content prompt word information set. Finally, the selected content prompt word information set and the tumor guideline drug analysis information set are input into the guideline content extraction large language model to obtain the drug recommendation information triple set.

[0076] In the fifth step, the drug recommendation information triple set is input into a preset database to obtain a tumor drug knowledge base. The preset database can be a pre-determined database. For example, the preset database can be a graph database.

[0077] The oncology drug retrieval server 105 is used to generate an oncology guideline recommended drug information set based on the oncology guideline drug recommendation query information set input by the user interactive terminal and the oncology drug knowledge base.

[0078] In some embodiments, the aforementioned executing entity can generate a set of recommended oncology guidelines for medication based on the set of query statements for oncology guidelines input by the user interaction terminal and the aforementioned oncology medication knowledge base. This recommended oncology guidelines for medication may be drug information that has been retrieved from the oncology medication knowledge base using search criteria and output in a structured format.

[0079] In some optional implementations of some embodiments, the aforementioned tumor drug retrieval server is configured as follows:

[0080] The first step involves determining that the query information for oncology guideline drug recommendations is oncology drug information. Based on this query information, a fuzzy matching process is performed on the oncology drug knowledge base to obtain a target set of oncology drug recommendations. This target set of oncology drug recommendations can be drug information generated solely from key oncology drug information included in the query information from oncology guideline drug recommendations.

[0081] The second step involves filtering the aforementioned target cancer type drug recommendation information set in response to the detected selection operation, thereby obtaining cancer guideline-recommended drug information. The selection operation for the cancer type can be performed by the user on various cancer types listed in the cancer guideline drug information generation system. For example... Figure 2 As shown, Figure 2 This diagram illustrates the oncology guidelines recommending medications when a user selects a type of cancer and enters the name of a cancer drug. Figure 2 The upper section displays the cancer types and drug names available for user selection. Figure 2 The lower half displays the recommended medication information from oncology guidelines extracted based on the search criteria.

[0082] The third step involves filtering the target oncology drug recommendation information set in response to the detected selection operation based on patient population stratification information, thereby obtaining oncology guideline-recommended drug information. The aforementioned patient population stratification information can be selected by the user based on the patient's specific cancer type and disease status. For example... Figure 3 As shown, Figure 3 This diagram illustrates the searchable criteria for oncology guideline medication information, tailored to patient population stratification, available to selectable users within the non-small cell lung cancer category.

[0083] In some optional implementations of some embodiments, the aforementioned tumor drug retrieval server is configured as follows:

[0084] The first step is to perform the following fuzzy matching steps for each oncology guideline drug recommendation query in the above-mentioned oncology guideline drug recommendation query information set:

[0085] Sub-step 1 involves determining the part-of-speech tag group for the medication recommendation query segmentation group corresponding to the aforementioned oncology guideline medication recommendation query information. The medication recommendation query segments in this group can be words related to the medication recommendations in the oncology guideline medication information, obtained by segmenting the oncology guideline medication recommendation query information using a word segmentation algorithm. The part-of-speech tags of the medication recommendation segments in this group can characterize the part-of-speech of the medication recommendation query segments. In practice, the executing entity can first use a word segmentation algorithm to segment the aforementioned oncology guideline medication recommendation query information to obtain the medication recommendation query segmentation group. Then, the part-of-speech tags of the medication recommendation query segmentation group are determined as the part-of-speech tags of the medication recommendations, thus obtaining the medication recommendation segmentation tag group.

[0086] Sub-step 2 involves performing entity boundary detection on the aforementioned medication recommendation query segmentation group based on the part-of-speech tags of the medication recommendation segmentation group, resulting in a medication segmentation boundary feature vector group. The medication segmentation boundary feature vectors in this group represent the positional information of each individual Chinese character within the medication recommendation query segmentation. For example, the execution entity can first use the GloVe (Global Vectors for Word Representation) model to represent the medication recommendation query segmentation group as word vectors, obtaining a medication recommendation query vector set. Then, it encodes the part-of-speech tags of the medication recommendation segmentation group, obtaining a medication recommendation query part-of-speech vector set. Next, it concatenates the medication recommendation query vector set and the medication recommendation query part-of-speech vector set to obtain a medication recommendation query vector group. Finally, it inputs the medication recommendation query vector group into a BiGRU (Bidirectional Gated Recurrent Unit) model to obtain a medication recommendation segmentation semantic feature vector group. Among them, the semantic feature vector of the drug recommendation segmentation can represent the semantic, syntactic, and positional information of the drug recommendation query segmentation. Finally, using the BIMESO standard method, entity boundary detection is performed on the semantic feature vector group of the drug recommendation segmentation to obtain the boundary feature vector group of the drug segmentation.

[0087] Sub-step 3, the drug information query syntax dependency graph is obtained by performing syntax dependency analysis on the above-mentioned tumor guideline drug recommendation query sentence information. The drug information query syntax dependency graph can be a directed acyclic graph representing the dependency relationship between the drug recommendation query word groups in terms of grammar and semantics. The dependency relationship can include at least one of the following: subject-predicate dependency relationship, modifier dependency relationship of adverbial modifier of predicate, and verb-object dependency relationship.

[0088] Sub-step 4, the drug recommendation query information set is obtained by predicting the word segmentation dependency relationship of the drug recommendation query word groups according to the drug information query syntax dependency graph and the drug word segmentation boundary feature vector set. The drug recommendation query information in the drug recommendation query information set can be query information representing the association topological relationship between the drug recommendation query word groups in the form of a triple.

[0089] As an example, the execution subject can first convert the drug information query syntax dependency graph into a drug information query syntax dependency matrix. Then, the drug information query syntax dependency matrix and the drug word segmentation boundary feature vector set are input into the graph attention network to obtain a query sentence sequence feature vector set and a query word segmentation dependency feature vector set. The query sentence sequence feature vector in the query sentence sequence feature vector set can represent the sentence sequence information of each drug recommendation query word in the tumor guideline drug recommendation query sentence information. The query word segmentation dependency feature vector in the query word segmentation dependency feature vector set can represent the local dependency information between the drug recommendation query word groups. Finally, the query sentence sequence feature vector set and the query word segmentation dependency feature vector set are input into the sigmoid activation function layer for word segmentation relationship prediction to obtain the drug recommendation query information set.

[0090] Sub-step 5, the drug recommendation query structured information set is obtained by performing entity linking mapping on the drug recommendation query information set. The drug recommendation query structured information in the drug recommendation query structured information set can be the information of drug treatment related to the tumor guideline drug information in the tumor guideline drug recommendation query sentence information, which is output in the form of json. For example, the drug recommendation query structured information can be {“non-small cell lung cancer”:{“treatment line”:[“first-line treatment”],“genotype”:[“EGFR sensitive mutation”]}}. For example, Figure 4As shown, the user inputs "EGFR (Epidermal Growth Factor Receptor) sensitive mutation non-small cell tumor treatment options" under the natural language retrieval option, and the large model engine generates drug recommendation query structured data and retrieves recommended information through the drug recommendation query structured data to obtain a page diagram of the recommended information.

[0091] In sub-step 6, the recommended fuzzy matching of the tumor drug knowledge base is performed according to the above-mentioned drug recommendation query structured information set, and tumor guideline drug information is obtained. As an example, the execution subject can use a fuzzy matching algorithm to perform recommended fuzzy matching of the tumor drug knowledge base according to the above-mentioned drug recommendation query structured information set, and obtain tumor guideline recommended drug information.

[0092] The tumor drug retrieval server 105 is used to send the tumor guideline recommended drug information set to the user interactive terminal.

[0093] In some embodiments, the execution subject can send the tumor guideline recommended drug information set to the user interactive terminal. The user interactive terminal can be a terminal that receives the tumor guideline recommended drug information set sent by the tumor drug retrieval server and displays the information on the terminal for medical personnel to use.

[0094] The user interactive terminal 106 is used to display the tumor guideline recommended drug information set on the terminal in the form of voice or text.

[0095] In some embodiments, the execution subject can display the tumor guideline recommended drug information set on the terminal in the form of voice or text.

[0096] In addition, it should be noted that the tumor guideline drug information generation can also include a tumor knowledge base update server for periodically updating the tumor drug knowledge base according to a preset collection period. The preset collection period can be a preset period, and the tumor guideline drug information collection period can be set every preset time interval. The preset time interval can be determined according to specific conditions and is not limited herein. The update and maintenance can be to periodically collect the latest tumor guideline drug information set related to tumor information according to the preset collection period, and to analyze the latest tumor guideline drug information set and perform incremental difference update of the tumor drug database.

[0097] With reference to Figure 5 , Figure 5 Figure 500 shows the flow 500 of some embodiments of the tumor guideline drug information generation method according to the present disclosure. The tumor guideline drug information generation method includes the following steps:

[0098] In step 501, the data collection server acquires the tumor guideline drug information set and the tumor guideline drug recommendation query sentence information set.

[0099] In some embodiments, the above-mentioned tumor guideline drug information generation method can acquire the tumor guideline drug information set and the tumor guideline drug recommendation query sentence information set by the data collection server.

[0100] In step 502, the tumor guideline drug recommendation query sentence information set is classified and processed to obtain a tumor query sentence grouping information set.

[0101] In some embodiments, the above-mentioned execution subject can classify and process the above-mentioned tumor guideline drug recommendation query sentence information set to obtain a tumor query sentence grouping information set.

[0102] In step 503, according to the tumor query sentence grouping information set, a content extraction prompt word information set for the tumor guideline drug information set is generated.

[0103] In some embodiments, the above-mentioned execution subject can generate a content extraction prompt word information set for the tumor guideline drug information set according to the above-mentioned tumor query sentence grouping information set.

[0104] In step 504, according to the content extraction prompt word information set, the tumor guideline drug information set is subjected to information parsing processing to obtain a tumor guideline drug parsing information set.

[0105] In some embodiments, the above-mentioned execution subject can perform information parsing processing on the above-mentioned tumor guideline drug information set according to the above-mentioned content extraction prompt word information set to obtain a tumor guideline drug parsing information set.

[0106] In step 505, according to the content extraction prompt word information set and the guideline content extraction large language model, the tumor guideline drug parsing information set is subjected to drug knowledge extraction processing to obtain a tumor drug knowledge base.

[0107] In some embodiments, the above-mentioned execution subject can perform drug knowledge extraction processing on the above-mentioned tumor guideline drug parsing information set according to the above-mentioned content extraction prompt word information set and the guideline content extraction large language model to obtain a tumor drug knowledge base.

[0108] In step 506, according to the tumor guideline drug recommendation query sentence information set input by the user interaction terminal and the tumor drug knowledge base, a tumor guideline recommended drug information set is generated.

[0109] In some embodiments, the above-mentioned execution subject can generate a tumor guideline recommended drug information set according to the tumor guideline drug recommendation query sentence information set input by the above-mentioned user interaction terminal and the above-mentioned tumor drug knowledge base.

[0110] Step 507, send the tumor guideline recommended drug information set to the user interactive terminal.

[0111] In some embodiments, the execution subject can send the tumor guideline recommended drug information set to the user interactive terminal.

[0112] Step 508, display the tumor guideline recommended drug information set on the terminal in the form of voice or text.

[0113] In some embodiments, the execution subject can display the tumor guideline recommended drug information set on the terminal in the form of voice or text.

[0114] Reference will now be made to the following description Figure 6 , which shows a structural schematic diagram of an electronic device (e.g., electronic device) 600 suitable for use to implement some embodiments of the present disclosure. Figure 6 The electronic device shown is merely an example and should not bring any limitation to the function and use range of embodiments of the present disclosure.

[0115] As shown in Figure 6 , the electronic device 600 can include a processing device (e.g., central processor, graphics processor, etc.) 601, which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 602 or loaded from a storage device 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the electronic device 600 are also stored. The processing device 601, the ROM 602, and the RAM 603 are connected to each other through a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0116] Generally, the following devices can be connected to the I / O interface 605: input devices 606 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 608 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 609. The communication devices 609 can allow the electronic device 600 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 6 The electronic device 600 is shown with various devices, but it should be understood that all the devices shown are not required to be implemented or possessed. More or fewer devices can be alternatively implemented or possessed. Figure 6 Each block shown in the

[0117] In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program according to some embodiments of the present disclosure. For example, some embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer readable medium, the computer program comprising program code for performing the methods illustrated by the flowcharts. In some such embodiments, the computer program can be downloaded and installed from a network via the communication device 609, or installed from the storage device 608, or installed from the ROM 602. When the computer program is executed by the processing device 601, the above-mentioned functions defined in the methods of some embodiments of the present disclosure are performed.

[0118] It should be noted that the computer readable medium mentioned above in some embodiments of the present disclosure can be a computer readable signal medium or a computer readable storage medium or any combination of the two. The computer readable storage medium may, for example, be, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination of the above. More specific examples of the computer readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, the computer readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus or device. In some embodiments of the present disclosure, the computer readable signal medium can include a data signal carried in a baseband or as part of a carrier wave, in which the computer readable program code is carried. Such a propagated data signal can take many forms, including but not limited to, an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer readable signal medium can also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate or transport a program for use by or in connection with an instruction execution system, apparatus or device. The program code contained on the computer readable medium can be transmitted by any suitable medium, including but not limited to, wire, cable, RF (radio frequency), etc., or any suitable combination of the above.

[0119] In some embodiments, the client, server, or other computing devices can communicate information using any known or future developed end-to-end communications protocol, such as the Hyper Text Transfer Protocol (HTTP), and can be interconnected via any form or medium of digital data communication (for example, a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), the Internet, and peer-to-peer networks (for example, ad hoc peer-to-peer networks), as well as any current or future developed network.

[0120] The computer-readable medium can be included within the electronic device; alternatively, the computer-readable medium can exist solely within the electronic device, or can exist solely outside of the electronic device. The computer-readable medium can carry one or more programs, which, when executed by the electronic device, cause the electronic device to perform the steps described above.

[0121] Computer program code for carrying out operations of some embodiments of the present disclosure can be written in any of one or more programming languages, including object oriented programming languages such as Java, Smalltalk, C++, or conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network ("LAN") or a wide area network ("WAN"), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0122] The computer program product of the first aspect can further include one or more of the following features. The computer program product can include a computer readable medium. The computer readable medium can include a computer readable storage medium or a computer readable signal medium. The computer readable storage medium can include tangible storage medium. The computer readable signal medium can include a propagated data signal with computer readable program code embodied therein. The computer readable program code can be downloaded into a working memory of a computer from the computer readable signal medium or from the computer readable storage medium. The computer readable program code can cause the computer to perform the steps of the first aspect. The computer readable program code can be executed by one or more processors associated with the computer.

[0123] The functions described above can be performed, at least in part, by one or more hardware logic components. For example, and without limitation, non-limiting examples of hardware logic components that can be used include field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SOCs), complex programmable logic devices (CPLDs), etc.

[0124] The above description is merely exemplary of the disclosure and the application of the principles thereof. It is not intended to limit the scope of the disclosure to the precise forms disclosed. The disclosure is to cover all modifications, equivalents, and alternatives falling within the scope of the application as defined by the claims. The disclosure can also cover combinations of the features set out above.

Claims

1. A system for generating tumor guideline drug information, comprising a data collection server, a prompt word generation server, a drug information analysis server, a tumor drug database construction server, a tumor drug search server and a user interaction terminal, wherein: the data collection server is configured to obtain a tumor guideline drug information set and a tumor guideline drug recommendation query sentence information set; the tumor guideline drug recommendation query sentence information set is classified to obtain a tumor query sentence grouping information set; the prompt word generation server is configured to generate a content extraction prompt word information set for the tumor guideline drug information set according to the tumor query sentence grouping information set; the drug information analysis server is configured to perform information analysis processing on the tumor guideline drug information set according to the content extraction prompt word information set to obtain a tumor guideline drug analysis information set; the tumor drug database construction server is configured to perform drug knowledge extraction processing on the tumor guideline drug analysis information set according to the content extraction prompt word information set and a guideline content extraction large language model to obtain a tumor drug knowledge base; the tumor drug search server is configured to generate a tumor guideline recommended drug information set according to a tumor guideline drug recommendation query sentence information set input by the user interaction terminal and the tumor drug knowledge base; the tumor guideline recommended drug information set is sent to the user interaction terminal; and the user interaction terminal is configured to display the tumor guideline recommended drug information set in the form of voice or text. The drug information analysis server is configured to: perform block processing on the tumor guideline drug information set to obtain a tumor drug text information set, a tumor drug table set and a tumor drug flowchart set; perform character recognition on the tumor drug text information set to obtain a tumor drug text character information set; perform table character recognition on the tumor drug table set according to a table analysis prompt word information set included in the content extraction prompt word information set to obtain a tumor drug table information set; perform flowchart character recognition on the tumor drug flowchart set according to a flowchart analysis prompt word information set included in the content extraction prompt word information set to obtain a tumor drug flowchart information set; perform text correction processing on the tumor drug text character information set, the tumor drug table information set and the tumor drug flowchart information set to obtain a corrected tumor drug text character information set, a corrected tumor drug table information set and a corrected tumor drug flowchart information set; and determine the corrected tumor drug text character information set, the corrected tumor drug table information set and the corrected tumor drug flowchart information set as the tumor guideline drug analysis information set. The drug information analysis server is configured to: determine a tumor drug search term set according to the corrected tumor drug text character information set, the corrected tumor drug table information set and the corrected tumor drug flowchart information set included in the tumor guideline drug analysis information set; and perform semantic ambiguity resolution processing on the tumor drug search term set to obtain a disambiguated tumor drug search term set. ​ ​ ​ ​ ​ ​ 2. The tumor guide dosing information generation system according to claim 1, wherein ​ ​ ​ ​ ​ ​ ​ 3. The tumor guide dosing information generation system according to claim 1, wherein ​ ​ ​ The content extraction prompt word information set is subjected to prompt word optimization processing to obtain an optimized content prompt word information set; Through the guideline content extraction large language model, according to the optimized content prompt word information set and the disambiguated tumor drug search term set, the corrected tumor drug text information set, the corrected tumor drug table information set and the corrected tumor drug process information set are subjected to multiple-choice question form drug recommendation information extraction to obtain a drug recommendation information triple set; The drug recommendation information triple set is input into a preset database to obtain a tumor drug knowledge base.

4. The tumor guide dosing information generation system according to claim 1, wherein The tumor drug search server is configured to: For each tumor guideline drug recommendation query sentence information in the tumor guideline drug recommendation query sentence information set, the following recommendation fuzzy matching steps are performed: Determine the drug recommendation part-of-speech group of the drug recommendation query part-of-speech group corresponding to the tumor guideline drug recommendation query sentence information; According to the drug recommendation part-of-speech group, the drug recommendation query part-of-speech group is subjected to entity boundary detection to obtain a drug part-of-speech boundary feature vector group; The tumor guideline drug recommendation query sentence information is subjected to sentence dependency syntax analysis to obtain a query syntax dependency graph; According to the query syntax dependency graph and the drug part-of-speech boundary feature vector group, the drug recommendation query part-of-speech group is subjected to part-of-speech dependency relationship prediction to obtain a drug recommendation query information set; The drug recommendation query information set is subjected to part-of-speech entity linking mapping to obtain a drug recommendation query structured information set; According to the drug recommendation query structured information set, the tumor drug knowledge base is subjected to recommendation fuzzy matching to obtain tumor guideline recommended drug information.

5. The tumor guide dosing information generation system according to claim 4, wherein The tumor drug search server is configured to: In response to determining that the tumor guideline drug recommendation query sentence information is tumor drug information, according to the tumor guideline drug recommendation query sentence information, the tumor drug knowledge base is subjected to recommendation fuzzy matching to obtain a target tumor drug recommendation information set; In response to detecting a selection operation for a tumor cancer type, the target tumor drug recommendation information set is subjected to screening processing to obtain tumor guideline recommended drug information; In response to detecting a selection operation for patient population classification information, the target tumor drug recommendation information set is subjected to screening processing to obtain tumor guideline recommended drug information.

6. The tumor guide dosing information generation system according to claim 1, wherein The prompt word generation server is configured to: Determine a set of thought chain query decomposition step groups of the set of tumor query sentence grouping information groups; Determine the prompt word category information of each tumor query sentence grouping information group in the set of tumor query sentence grouping information groups; The prompt word category information, the tumor guideline drug information set and the set of thought chain query decomposition step groups are input into a prompt word generation model to obtain a content extraction prompt word information set.

7. A tumor guideline drug information generation method, applied to the tumor guideline drug information generation system of any one of claims 1-6, comprising: Obtaining a tumor guideline drug information set and a tumor guideline drug recommendation query sentence information set; Classifying the tumor guideline drug recommendation query sentence information set to obtain a set of tumor query sentence grouping information groups; According to the tumor query sentence grouping information set, a content extraction prompt word information set for the tumor guideline drug information set is generated; According to the content extraction prompt word information set, information analysis processing is performed on the tumor guideline drug information set to obtain a tumor guideline drug analysis information set; According to the content extraction prompt word information set and the guideline content extraction large language model, drug knowledge extraction processing is performed on the tumor guideline drug analysis information set to obtain a tumor drug knowledge base; According to the tumor guideline drug recommendation query sentence information set input by the user interaction terminal and the tumor drug knowledge base, a tumor guideline recommended drug information set is generated; The tumor guideline recommended drug information set is sent to the user interaction terminal; The tumor guideline recommended drug information set is displayed on the terminal in the form of voice or text.

8. An electronic device, comprising: one or more processors; a storage device having stored thereon one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 7.

9. A computer readable medium having stored thereon a computer program, wherein, The computer program is executed by the processor to implement the method of any one of claims 7.

Images