Is there a way to explore the topic recommendation analysis?

[j_15]

We ran the topic recommendation analysis on 70,000 incidents and got a multitude of recommendations. 

Before I select and publish the recommended topics, I'd like to verify that the topics that were deemed matches by the topic recommender, are actually matches.

Is there a way to do this?

[VarunaS]

Hi,

• Yes, you can verify the topics matched by the topic recommender.
• Create a script to retrieve the recommended topics from the analysis results.
• Compare the recommended topics with the incident descriptions.
• Use a script to automate the comparison for accuracy.
• Here’s a basic script snippet to get you started:
  

  // Loop through the incidents
  while (incidentsGR.next()) {
  // Retrieve the recommended topic from your analysis results
  var recommendedTopic = getRecommendedTopic(incidentsGR.number); // Implement getRecommendedTopic function
  
  // Verify if the recommended topic matches the incident description
  if (isTopicMatch(incidentsGR.short_description, recommendedTopic)) { // Implement isTopicMatch function
  matchedTopics.push({
  incidentNumber: incidentsGR.number,
  description: incidentsGR.short_description,
  recommendedTopic: recommendedTopic
  });
  }
  }

  // Function to get the recommended topic
  function getRecommendedTopic(incidentNumber) {
  // Retrieve the topic based on incident number (dummy implementation)
  // Replace this with actual retrieval logic from your analysis results
  return "Sample Topic";
  }

  // Function to check if the topic matches the description
  function isTopicMatch(description, topic) {
  // Implement your matching logic here
  // For example, use a keyword search or a similarity algorithm
  return description.includes(topic);
  }

  // Output the matched topics for review
  gs.info(JSON.stringify(matchedTopics));

• Adjust the getRecommendedTopic and isTopicMatch functions to fit your analysis results and matching criteria.
• Review the output to ensure the topics match the incidents accurately.
• Iterate and refine the matching logic as needed.
• Once verified, proceed to select and publish the recommended topics.
• Keep the process iterative and involve stakeholders for validation if needed.