/**
* Topic clustering for conversations
* Groups conversations by detected themes using keyword extraction
*/
/**
* Pre-defined topic categories with associated keywords
* Single regex per topic for simplicity - add more patterns by extending the alternation
*/
const TOPIC_PATTERNS: Record<string, RegExp> = {
'authentication': /\b(auth|login|logout|signin|signout|password|jwt|token|oauth|session|2fa|mfa)\b/i,
'database': /\b(database|sql|query|postgresql|postgres|mysql|mongodb|sqlite|prisma|drizzle|orm|migration|schema)\b/i,
'api': /\b(api|rest|graphql|endpoint|route|fetch|axios|request|response|http|webhook)\b/i,
'frontend': /\b(react|vue|angular|svelte|nextjs|component|jsx|tsx|css|tailwind|styling|ui|ux)\b/i,
'testing': /\b(test|testing|jest|vitest|mocha|cypress|playwright|spec|mock|fixture|coverage)\b/i,
'deployment': /\b(deploy|deployment|docker|kubernetes|k8s|ci|cd|pipeline|vercel|aws|gcp|azure|hosting)\b/i,
'performance': /\b(performance|optimize|optimization|cache|caching|lazy|bundle|memory|speed|slow|fast)\b/i,
'debugging': /\b(debug|error|exception|crash|bug|fix|issue|problem|trace|stack|log)\b/i,
'refactoring': /\b(refactor|restructure|cleanup|clean up|improve|simplify|extract|rename|move)\b/i,
'configuration': /\b(config|configuration|setup|install|env|environment|settings|options)\b/i,
'typescript': /\b(typescript|ts|type|interface|generic|typing|typed)\b/i,
'git': /\b(git|commit|merge|branch|rebase|push|pull|conflict|diff)\b/i,
'security': /\b(security|secure|vulnerability|xss|csrf|injection|sanitize|encrypt|decrypt)\b/i,
'state-management': /\b(state|redux|zustand|recoil|mobx|context|provider|store)\b/i,
};
export interface TopicMatch {
topic: string;
score: number; // Number of pattern matches
matchedTerms: string[];
}
export interface TopicCluster {
topic: string;
conversationCount: number;
conversationIds: string[];
keyTerms: string[];
}
/**
* Detect topics in a text
*/
export function detectTopics(text: string): TopicMatch[] {
const matches: TopicMatch[] = [];
for (const [topic, pattern] of Object.entries(TOPIC_PATTERNS)) {
const globalPattern = new RegExp(pattern.source, 'gi');
const textMatches = text.match(globalPattern);
if (textMatches && textMatches.length > 0) {
const uniqueTerms = [...new Set(textMatches.map(m => m.toLowerCase()))];
matches.push({
topic,
score: textMatches.length,
matchedTerms: uniqueTerms,
});
}
}
// Sort by score descending
matches.sort((a, b) => b.score - a.score);
return matches;
}
/**
* Get primary topic for a conversation
*/
export function getPrimaryTopic(text: string): string {
const topics = detectTopics(text);
return topics.length > 0 ? topics[0].topic : 'general';
}
/**
* Cluster conversations by topic
*/
export function clusterByTopic(
conversations: Array<{
conversationId: string;
searchableText: string;
}>
): TopicCluster[] {
const clusterMap = new Map<string, {
conversationIds: string[];
allTerms: string[];
}>();
for (const conv of conversations) {
const topics = detectTopics(conv.searchableText);
const primaryTopic = topics.length > 0 ? topics[0].topic : 'general';
if (!clusterMap.has(primaryTopic)) {
clusterMap.set(primaryTopic, { conversationIds: [], allTerms: [] });
}
const cluster = clusterMap.get(primaryTopic)!;
cluster.conversationIds.push(conv.conversationId);
// Collect top matched terms
if (topics.length > 0) {
cluster.allTerms.push(...topics[0].matchedTerms);
}
}
// Convert to array and compute key terms
const clusters: TopicCluster[] = [];
for (const [topic, data] of clusterMap) {
// Count term frequency
const termCounts = new Map<string, number>();
for (const term of data.allTerms) {
termCounts.set(term, (termCounts.get(term) || 0) + 1);
}
// Get top 5 terms by frequency
const keyTerms = [...termCounts.entries()]
.sort((a, b) => b[1] - a[1])
.slice(0, 5)
.map(([term]) => term);
clusters.push({
topic,
conversationCount: data.conversationIds.length,
conversationIds: data.conversationIds,
keyTerms,
});
}
// Sort by conversation count descending
clusters.sort((a, b) => b.conversationCount - a.conversationCount);
return clusters;
}
/**
* Get topic distribution for conversations
*/
export function getTopicDistribution(
conversations: Array<{
conversationId: string;
searchableText: string;
}>
): Record<string, number> {
const distribution: Record<string, number> = {};
for (const conv of conversations) {
const topics = detectTopics(conv.searchableText);
const primaryTopic = topics.length > 0 ? topics[0].topic : 'general';
distribution[primaryTopic] = (distribution[primaryTopic] || 0) + 1;
}
return distribution;
}
