Intelligent Agent memory field knowledge conflict detection method
Through intelligent Agent detection and update domain knowledge conflicts, the problem of inconsistency in large models in long conversations is solved, the user experience and answer coherence is improved, and the accuracy and consistency of the knowledge base is ensured.
Patent Information
- Application Number
- CN202510497870.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-21
- Publication Date
- 2025-09-05
AI Technical Summary
In long conversation scenarios, large models tend to forget the information before and after or provide inconsistent answers, resulting in a decline in user experience and a decrease in trust. The information provided by users is vague and does not have a consistent structure, which increases the difficulty of generating accurate answers.
Through the intelligent Agent memory domain knowledge conflict detection method, we detect whether the information input by the user is a problem or domain knowledge, and find similar knowledge in memory, determine whether it is a contradiction, update the knowledge base, ensure the consistency of knowledge, use the embedding algorithm to calculate the sentence similarity and judge the contradiction through a big model, prompting the user to handle the conflict.
Improves the memory ability and consistency of large models in long conversations, improves user experience, and ensures the accuracy and consistency of domain knowledge.
Smart Images
Figure CN120596598A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of artificial intelligence large models, and in particular is a method for detecting knowledge conflicts in intelligent agent memory domains. Background Art
[0002] Currently, traditional large models rely on the Transformer architecture. While this architecture performs well when processing short-term information, it often forgets previous information or provides inconsistent responses in extended conversations. This phenomenon is primarily due to the limited context window of large models, which cannot retain the user's previous input and background information over the long term. If the model fails to maintain consistency during extended conversations with intelligent agents, it can lead to a poor user experience and reduced trust in the model's generated answers. To address this issue, researchers have proposed a solution that incorporates long-term memory mechanisms into intelligent agents. By incorporating long-term memory, intelligent agents can retain and retrieve previous user information during conversations, thereby improving the model's contextual understanding and consistent responses. However, simply extending the memory range of large models is not sufficient to solve all problems. User questions are often vague and lack a consistent structure, which makes it difficult to generate accurate responses.
[0003] Large models and AI agent technologies are increasingly being used in modern artificial intelligence. Their respective advantages make them a crucial force driving technological advancement and industry transformation. The power of large models stems from their deep understanding of natural language processing. These models, such as GPT and BERT, are trained on massive amounts of text data, enabling them to understand context, identify language patterns, and generate high-quality text. Their self-attention mechanism and Transformer architecture make large models particularly effective when processing long texts and complex semantics. This capability makes large models highly practical and accurate in a variety of fields, such as machine translation, text generation, summarization, and sentiment analysis. Compared to traditional models, the universality and flexibility of large models enable them to quickly adapt to a wide range of application scenarios, thereby improving development and application efficiency. AI agent technology further advances the autonomous decision-making capabilities of intelligent systems. AI agents are capable of real-time learning and adaptation in dynamic and complex environments. They continuously optimize their decision-making processes through reinforcement learning and other machine learning algorithms. AI agents not only perform tasks but also continuously adjust their strategies based on feedback during execution. This self-learning capability enables AI agents to enhance automation and user experience in many practical applications. The combination of large models and AI agents further enhances the overall intelligence of intelligent systems. Large models provide AI agents with powerful language understanding and generation capabilities, making human-computer interaction more natural and fluid. This combination not only enhances the user-system interaction experience but also provides AI agents with a rich knowledge base, enabling them to make more accurate decisions when faced with complex problems. The paper [Understanding the benefits and challenges of using large language model-based conversational agents for mental well-being support] proposes a new conversational agent. During communication, the researcher's agent can provide useful feedback to guide the decision-making agent's actions. However, general large models struggle to adapt to users' specialized industry needs, and ordinary users lack the hardware and skills to fine-tune large models. Leveraging the long-term memory capabilities of intelligent agents can significantly enhance large models' ability to handle specialized problems at minimal cost. However, large models are highly sensitive to user prompts, and prompts that provide more useful information can significantly improve the accuracy and usefulness of answers. Using higher-quality prompts to supplement or replace user questions can significantly improve user satisfaction.The paper [Automatic Engineering of Long Prompts] found that greedy algorithms and genetic algorithms perform well in constructing long prompts, and that long prompts significantly improve accuracy for tasks in BigBenchHard. However, when seeking help from intelligent agents, users generally do not provide many prompts. Instead, they tend to ask multiple questions in short sentences, which provides very limited information and sometimes contradicts each other. This poses a significant challenge for agents to accurately answer user questions. Summary of the Invention
[0004] The purpose of the present invention is to provide a method for detecting conflicts in domain knowledge memorized by an intelligent agent, which can detect whether domain knowledge statements in the memory of the same user agent or knowledge in different knowledge bases are contradictory, or detect whether domain knowledge statements in different user agents within the same organization are contradictory and submit them to the user for processing, thereby effectively ensuring the consistency of domain knowledge.
[0005] The purpose of the present invention is achieved through the following solutions:
[0006] A method for detecting conflicts in domain knowledge memory of an intelligent agent, characterized by comprising the following steps:
[0007] 1) Determine the user input information and answer it directly if it is a question; otherwise, consider it as domain expertise that needs to be memorized and proceed to the next step;
[0008] 2) When it is considered as domain knowledge, the memory is searched for past knowledge similar to the current input. If found, it is determined whether the past knowledge contradicts the current knowledge. If so, the user is asked to select the correct knowledge and update the knowledge base;
[0009] 3) When a user raises a question and uploads it to a knowledge base, the system searches multiple knowledge bases, performs knowledge conflict detection on the search results, and prompts the user to resolve any conflicts.
[0010] 4) Detect the professional knowledge in the memory of different users of the same organization and prompt users to deal with contradictions when they occur.
[0011] The user input information is judged and answered directly if it is a question; otherwise, it is considered as domain expertise that needs to be memorized and the next processing steps are as follows:
[0012] Step 11: The agent receives the user's input.
[0013] Step 12 constructs the statement "Is the statement '${user input}' a knowledge statement?" and uses the agent's built-in model to determine whether the user input is a question or a statement of knowledge. ${user input} refers to the statement entered by the user.
[0014] Step 13: When the big model determines that the user input is a question, the agent directly uses the big model contained in the agent to answer and feed back the result to the user.
[0015] Otherwise, step 14 considers it as a domain personal knowledge modification that needs to be recorded and stored.
[0016] When it is regarded as domain knowledge, the memory is searched for past knowledge similar to the current input. If found, it is determined whether the past knowledge contradicts the current knowledge. If there is a contradiction, the user is allowed to select the correct knowledge and update the knowledge base. The steps are as follows:
[0017] Step 21: Enter the user input segment input and the past user input declarative knowledge segments stored in the Agent memory stored Divide by sentences, or by punctuation marks "." and "?", and delete sentences with question marks directly.
[0018] Step 22 uses the embedding algorithm to obtain the embedding vector v for each sentence. Then calculate the segment input and segment stored The similarity between each embedding vector in
[0019]
[0020] Step 23 and The corresponding sentence groups are and Take the two groups of sentences with the greatest similarity, each of which contains a user input segment input The sentence in the agent and the past knowledge segment stored in the agent stored The sentences in .
[0021] Step 24: When their similarity is greater than or equal to the threshold T, further construct a question sentence in the following form to determine whether they are contradictory: "statement" "AND Statement" "Are they contradictory?"
[0022] Step 25 uses regularization rules to determine whether the answers from the large model are contradictory. If they are contradictory, the user is prompted to select the correct knowledge, and the incorrect knowledge is deleted from the user agent's memory, leaving only the correct knowledge.
[0023] When a user raises a question and uploads it to a knowledge base, a search is performed in multiple knowledge bases, and a knowledge conflict detection is performed on the search results. When they conflict with each other, the user is prompted to handle the following steps:
[0024] Step 31: segment the current question raised by the user input , search in multiple knowledge bases uploaded by users, and use the large model inside the intelligent agent to directly answer when no relevant information is found.
[0025] Step 32: When relevant information is found, suppose multiple relevant results are retrieved: answer1, answer2, ...answer n , combine these results into a long sentence answer con =answer1+answer2+...answer n , construct the sentence 'Are there any contradictions in the following statements: ${answer con}' and submit the large model solution;
[0026] Step 33: If the big model finds no contradiction, it generates the answer and directly submits it to the user. Otherwise, it uses the embedding algorithm to obtain the embedding vectors v1, v2, ...v of these results. n And calculate the similarity between them, and find the vectors whose similarity is greater than or equal to the threshold T;
[0027] Step 34 uses the aforementioned method to determine whether these vectors are contradictory. If they are contradictory, the user is prompted to handle the problem.
[0028] The aforementioned detection of professional knowledge among different users of the same organization and prompting the user to handle conflicts when they occur are as follows:
[0029] Step 41 sequentially takes a user u i and its knowledge base KL(u i )={v1,v2,...v M};
[0030] Step 42: i+1 Start by taking a user u j and its knowledge base KL(u j )={v'1,v'2,...v' N};
[0031] Step 43 From KL (u i ) in sequence to take a vector v p (i)1<=i<=M, calculate its difference with KL(u j ) in any vector v' q (j) 1 <= q <= N similarity, and obtain v p (i) with v' q (j)∈KL(u j ) is the maximum value of the similarity, that is
[0032]
[0033] Step 44 When simi(v p (i),v' q (j))>T when constructing statements to determine their corresponding statements s p (i),s' q (j) whether they are contradictory. If they are contradictory, go to step 45; otherwise, go to step 43;
[0034] Step 45: Get v p (i) The corresponding sentence and its knowledge base KL(u i ) in the previous sentence. If they are in the same paragraph, combine them into one long sentence. p (i)={s p-1 (i),s p (i)}, otherwise ss p (i) = s p (i);
[0035] Step 46: Take v' q (j) The corresponding sentence and its knowledge base KL(u j ) in the previous sentence. If they are in the same paragraph, combine them into one long sentence ss' q (j) = {s q-1 (j),s q (j)}, otherwise ss' q (j) = s' q (j);
[0036] Step 47 Construct statement judgment statement ss p (i),ss' q (j) Are they contradictory? If so, prompt the user u i and u j to be processed.
[0037] The beneficial effects of the present invention are:
[0038] This method adjusts the user's question, obtains the original question and the rephrased answer, and uses the Mamba network to select the answer that best matches the user's intent. This approach not only improves the memory capacity of large models but also, by optimizing question structure, ensures that intelligent agents can provide more coherent responses when handling complex conversations. This addresses the problem of inconsistent information in large models when handling long conversations, while also improving the overall user experience. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 It is the overall step-by-step structural diagram of the present invention. DETAILED DESCRIPTION
[0040] Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent like or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present invention, as detailed in the appended claims.
[0041] A method for detecting conflicts in domain knowledge memory of an intelligent agent includes the following steps:
[0042] 1) Determine the user input information. If it is a question, answer it directly. Otherwise, treat it as domain expertise that needs to be memorized and proceed to the next step. The details are as follows:
[0043] Step 11: The agent receives the user's input.
[0044] Step 12 constructs the statement "Is the statement '${user input}' a knowledge statement?" and uses the agent's built-in model to determine whether the user input is a question or a statement of knowledge. ${user input} refers to the statement entered by the user.
[0045] Step 13: When the big model determines that the user input is a question, the agent directly uses the big model contained in the agent to answer and feed back the result to the user.
[0046] Otherwise, step 14 considers it as a domain personal knowledge modification that needs to be recorded and stored.
[0047] 2) When it is considered domain knowledge, the system searches for past knowledge similar to the current input in memory. If found, it determines whether the past knowledge contradicts the current knowledge. If so, it allows the user to select the correct knowledge and update the knowledge base. The details are as follows:
[0048] Step 21: Enter the user input segment inputand the past user input declarative knowledge segments stored in the Agent memory stored Divide by sentence, for example, you can divide by punctuation marks "." and "?", and delete the sentences with question marks directly.
[0049] Step 22 uses the embedding algorithm to obtain the embedding vector v for each sentence. Then calculate the segment input and segment stored The similarity between each embedding vector in
[0050]
[0051] Step 23 and The corresponding sentence groups are and Take the two groups of sentences with the greatest similarity, each of which contains a user input segment input The sentence in the agent and the past knowledge segment stored in the agent stored The sentences in .
[0052] Step 24: When their similarity is greater than or equal to the threshold T, further construct a question sentence in the following form to determine whether they are contradictory: "statement" "AND Statement" "Are they contradictory?"
[0053] Step 25 uses regularization rules to determine whether the answers from the large model are contradictory. If they are contradictory, the user is prompted to select the correct knowledge, and the incorrect knowledge is deleted from the user agent's memory, leaving only the correct knowledge.
[0054] 3) When a user asks a question and uploads it to a knowledge base, a search is performed in multiple knowledge bases, and the search results are checked for knowledge conflicts. When they conflict with each other, the user is prompted to handle the conflict. The details are as follows:
[0055] Step 31: segment the current question raised by the user input , search in multiple knowledge bases uploaded by users, and use the large model inside the intelligent agent to directly answer when no relevant information is found.
[0056] Step 32: When relevant information is found, suppose multiple relevant results are retrieved: answer1, answer2, ...answer n , combine these results into a long sentence answer con=answer1+answer2+...answer n , construct the sentence 'Are there any contradictions in the following statements: ${answer con}' and submit the large model answer
[0057] Step 33: If the big model finds no contradiction, it generates the answer and directly submits it to the user. Otherwise, it uses the embedding algorithm to obtain the embedding vectors v1, v2, ...v of these results. n And calculate the similarity between them, find the vectors whose similarity is greater than or equal to the threshold T
[0058] Step 34 uses the aforementioned method to determine whether these vectors are contradictory. If they are contradictory, the user is prompted to handle the problem.
[0059] 4) Detect the professional knowledge stored by different users of the same organization and prompt users to resolve any conflicts as follows:
[0060] Step 41 sequentially takes a user u i and its knowledge base KL(u i )={v1,v2,...v M};
[0061] Step 42: i+1 Start by taking a user u j and its knowledge base KL(u j )={v'1,v'2,...v' N};
[0062] Step 43 From KL (u i ) in sequence to take a vector v p (i)1<=i<=M, calculate its difference with KL(u j ) in any vector v' q (j) 1 <= q <= N similarity, and obtain v p (i) with v' q (j)∈KL(u j ) is the maximum value of the similarity, that is
[0063]
[0064] Step 44 When simi(v p (i),v' q (j))>T when constructing statements to determine their corresponding statements s p (i),s' q (j) whether they are contradictory. If they are contradictory, go to step 45; otherwise, go to step 43;
[0065] Step 45: Get v p (i) The corresponding sentence and its knowledge base KL(u i ) in the previous sentence. If they are in the same paragraph, combine them into one long sentence. p (i)={s p-1 (i),s p (i)}, otherwise ss p (i) = s p (i);
[0066] Step 46: Take v' q (j) The corresponding sentence and its knowledge base KL(u j ) in the previous sentence. If they are in the same paragraph, combine them into one long sentence ss' q (j) = {s q-1 (j),s q (j)}, otherwise ss' q (j) = s' q (j);
[0067] Step 47 Construct statement judgment statement ss p (i),ss' q (j) Are they contradictory? If so, prompt the user u i and u j to be processed.
[0068] The specific embodiments of the present invention are described in further detail below.
[0069] 1. User knowledge discovery.
[0070] The agent receives user input and constructs the statement "Is the statement '${user input}' a knowledge statement?" Using the agent's internal model, it determines whether the user input is a question or a statement of knowledge, where ${user input} refers to the statement entered by the user. If the model determines that the user input is a question, it directly uses the agent's internal model to provide an answer and feeds the result back to the user. Otherwise, it is considered a domain knowledge correction that needs to be recorded and stored by the agent.
[0071] For example, if the user inputs "Which pumped-storage power station currently has the highest rated head in the world?", the sentence "Is the statement 'Which pumped-storage power station currently has the highest rated head in the world?' a knowledge statement?" is constructed and submitted to the large model in the agent for answer. The large model's answer is as follows:
[0072]
[0073] Use the regular expression r'not[\u4e00-\u9fff]*knowledge statement' to determine that it is not a knowledge statement, then directly pass the user input to the large language model to obtain the result and return it to the user.
[0074] For example, if the user inputs "Currently, the pumped-storage power station with the highest rated water head in the world is the Tiantai Pumped-storage Power Station in Zhejiang, China.", a sentence is constructed: "Is the statement 'Currently, the pumped-storage power station with the highest rated water head in the world is the Tiantai Pumped-storage Power Station in Zhejiang, China.' a knowledge statement?" and submitted to the large model in the agent for answer. The large model's answer is as follows:
[0075]
[0076] Use the regular expression r'not[\u4e00-\u9fff]*knowledge statement' to determine whether it is a knowledge statement, which is a domain knowledge modification that needs to be recorded and stored by the Agent.
[0077] 2. Conflict detection and handling.
[0078] Enter user input into segment input and the past user input declarative knowledge segment stored in the Agent stored Divide by sentence, for example, by punctuation marks "." and "?", and delete the sentences with question marks. Use the embedding algorithm to get the embedding vector v for each sentence.
[0079] Then calculate the segment input and segment stored The similarity between each embedding vector in
[0080]
[0081] Suppose the sentence groups corresponding to and are and respectively. Take the two groups of sentences with the greatest similarity, each of which contains a user input segment input The sentence in the agent and the past knowledge segment stored in the agent stored When their similarity is greater than or equal to the threshold T, we further construct a question sentence in the following form to determine whether they are contradictory: "statement" "AND Statement" "Are they contradictory?"
[0082] Based on the answers from the large model, regular rules are used to determine whether they are contradictory. If they are contradictory, the user is prompted to select the correct knowledge, and the incorrect knowledge is deleted from the user agent's memory, leaving only the correct knowledge.
[0083] Suppose the user has previously entered "The pumped-storage power station with the highest rated water head in the world is the Katsuro River Pumped-Storage Power Station in Japan" and then entered "The pumped-storage power station with the highest rated water head in the world is the Tiantai Pumped-Storage Power Station in Zhejiang, China". The algorithm detects that these two pieces of knowledge are contradictory, and therefore prompts the user to choose the correct understanding.
[0084]
[0085] 3. Detect knowledge conflicts among multiple users within the same organization.
[0086] Sequentially take a user u i and its knowledge base KL(u i )={v1,v2,...v M}, from u i+1 Start by taking a user u j and its knowledge base KL(u j )={v'1,v'2,...v' N}
[0087] From KL i ) in sequence to take a vector v p (i)1<=i<=M, calculate its difference with KL(u j ) in any vector v' q (j) 1 <= q <= N similarity, and obtain v p (i) with v' q (j)∈KL(u j ) is the maximum value of the similarity, that is
[0088]
[0089] When simi(v p (i),v' q (j))>T when constructing statements to determine their corresponding statements s p (i),s' q (j) Whether they are contradictory. If so, go to the next step; otherwise, go to the previous step.
[0090] Take v p (i) The corresponding sentence and its knowledge base KL(u i ) in the previous sentence. If they are in the same paragraph, combine them into one long sentence. p (i)={sp-1 (i),s p (i)}, otherwise ss p (i) = s p (i)
[0091] Take v' q (j) The corresponding sentence and its knowledge base KL(u j ) in the previous sentence. If they are in the same paragraph, combine them into one long sentence ss' q (j) = {s q-1 (j),s q (j)}, otherwise ss' q (j) = s' q (j)
[0092] Construct statement judgment statement ss p (i),ss' q (j) Are they contradictory? If so, prompt the user u i and u j to be processed.
[0093] The present invention detects whether domain knowledge statements in the memory of the same user agent or knowledge in different knowledge bases are contradictory, or detects whether domain knowledge statements in different user agents within the same organization are contradictory and submits them to the users for processing, which can effectively ensure the consistency of domain knowledge.
Claims
1. A method for detecting conflicts in domain knowledge memory of an intelligent agent, characterized in that: The following steps are involved: 1) Determine the user input information and answer it directly if it is a question; otherwise, consider it as domain expertise that needs to be memorized and proceed to the next step; 2) When it is considered as domain knowledge, the memory is searched for past knowledge similar to the current input. If found, it is determined whether the past knowledge contradicts the current knowledge. If so, the user is asked to select the correct knowledge and update the knowledge base; 3) When a user raises a question and uploads it to a knowledge base, the system searches multiple knowledge bases, performs knowledge conflict detection on the search results, and prompts the user to resolve any conflicts. 4) Detect the professional knowledge in the memory of different users of the same organization and prompt users to deal with contradictions when they occur.
2. The method for detecting conflicts in domain knowledge memory of an intelligent agent according to claim 1, characterized in that: Step 1) is as follows: Step 11: The agent receives the user's input; Step 12: Construct the statement "Is the statement '${user input}' a knowledge statement?" and use the agent's built-in large model to determine whether the user input is a question or a statement of knowledge, where ${user input} refers to the statement entered by the user. Step 13: When the big model determines that the user input is a question, it directly uses the big model contained in the agent to answer and feed back the result to the user; Step 14: Otherwise, it is considered as a domain personal knowledge that needs to be recorded and stored.
3. The method for detecting conflicts in domain knowledge memory of an intelligent agent according to claim 1, characterized in that: Step 2) is as follows: Step 21: Enter user input into segment input and the past user input declarative knowledge segments stored in the Agent memory stored Divide by sentence, by punctuation marks "." and "?", and delete sentences with question marks directly; Step 22: Use the embedding algorithm to get the embedding vector v for each sentence, so Then calculate the segment input and segment stored The similarity between each embedding vector in Step 23, set and The corresponding sentence groups are and Take the two groups of sentences with the greatest similarity, each of which contains a user input segment input The sentence in the agent and the past knowledge segment stored in the agent stored The sentence in Step 24: When their similarity is greater than or equal to the threshold T, further construct a question sentence in the following form to determine whether they are contradictory: Step 25: Use regular rules to determine whether the answers from the large model are contradictory. If they are contradictory, prompt the user to select the correct knowledge, delete the incorrect knowledge from the user agent's memory, and only retain the correct knowledge.
4. The method for detecting conflicts in domain knowledge memory of an intelligent agent according to claim 1, characterized in that: Step 3) is as follows: Step 31: Segment the current question raised by the user input ,searches in multiple knowledge bases uploaded by users, and directly answers the questions using the large model inherent in the intelligent agent; Step 32: When relevant information is found, suppose multiple relevant results are retrieved: answer1, answer2, ...answer n , combine these results into a long sentence answer con =answer1+answer2+...answer n , construct the sentence 'Are there any contradictions in the following statements: ${answer con }' and submit the large model solution; Step 33: If the big model finds no contradiction, it generates the answer and directly submits it to the user. Otherwise, it uses the embedding algorithm to obtain the embedding vectors v1, v2, ...v of these results. n And calculate the similarity between them, and find the vectors whose similarity is greater than or equal to the threshold T; Step 34: Use the above method to determine whether these vectors are contradictory. If they are contradictory, prompt the user to handle the problem.
5. The method for detecting conflicts in domain knowledge memory of an intelligent agent according to claim 1, characterized in that: Step 4) is as follows: Step 41: Sequentially select a user u i and its knowledge base KL(u i )={v1,v2,...v M }; Step 42, from u i+1 Start by taking a user u j and its knowledge base KL(u j )={v'1,v'2,...v' N }; Step 43, from KL (u i ) in sequence to take a vector v p (i)1<=i<=M, calculate its difference with KL(u j ) in any vector v' q (j) 1 <= q <= N similarity, and obtain v p (i) with v' q (j)∈KL(u j ) is the maximum value of the similarity, that is Step 44, when simi(v p (i),v' q (j))>T when constructing statements to determine their corresponding statements s p (i),s' q (j) whether they are contradictory. If they are contradictory, go to step 45; otherwise, go to step 43; Step 45, take v p (i) The corresponding sentence and its knowledge base KL(u i ) in the previous sentence. If they are in the same paragraph, combine them into one long sentence. p (i)={s p-1 (i),s p (i)}, otherwise ss p (i) = s p (i); Step 46: Take v' q (j) The corresponding sentence and its knowledge base KL(u j ) in the previous sentence. If they are in the same paragraph, combine them into one long sentence ss' q (j) = {s q-1 (j),s q (j)}, otherwise ss' q (j) = s' q (j); Step 47: Construct statement judgment statement ss p (i),ss' q (j) Are they contradictory? If so, prompt the user u i and u j to be processed.