summaryExtractor.ts•7.15 kB
/**
* Summary Extraction Tool: Extract key information from conversation content
*
* This module provides functionality to extract key information from complete conversations, using multiple strategies to identify important content:
* 1. Keyword matching: Identify sentences containing specific keywords
* 2. Sentence importance scoring: Evaluate sentence importance based on factors such as position, length, etc.
* 3. Contextual relevance: Consider logical connections between sentences
*/
// Define keywords and their weights
const KEYWORDS = {
// Task related
task: 1.5,
feature: 1.3,
implement: 1.3,
develop: 1.3,
complete: 1.2,
execute: 1.2,
verify: 1.2,
error: 1.5,
problem: 1.5,
fix: 1.5,
failure: 1.8,
success: 1.5,
dependency: 1.2,
blocking: 1.4,
risk: 1.4,
optimize: 1.3,
improve: 1.3,
// Decision related
decide: 1.6,
choose: 1.5,
decision: 1.6,
solution: 1.5,
architecture: 1.5,
design: 1.4,
structure: 1.4,
// Technology related
code: 1.3,
test: 1.3,
function: 1.2,
interface: 1.2,
type: 1.2,
module: 1.2,
component: 1.2,
data: 1.3,
file: 1.2,
path: 1.1,
// System state
status: 1.3,
start: 1.3,
stop: 1.3,
create: 1.3,
delete: 1.4,
update: 1.3,
query: 1.2,
// Negative information (needs special attention)
warning: 1.8,
exception: 1.8,
crash: 2.0,
severe: 1.8,
danger: 1.8,
urgent: 1.9,
};
/**
* Extract a brief summary from text
* @param text Text from which to extract the summary
* @param maxLength Maximum length of the summary
* @returns Extracted summary text
*/
export function extractSummary(text: string, maxLength: number = 100): string {
if (!text) return "";
// Remove Markdown formatting
const plainText = text
.replace(/```[\s\S]*?```/g, "") // Remove code blocks
.replace(/#+\s/g, "") // Remove header markers
.replace(/\*\*/g, "") // Remove bold markers
.replace(/\*/g, "") // Remove italic markers
.replace(/\[([^\]]+)\]\([^)]+\)/g, "$1") // Replace links with text
.replace(/\n/g, " ") // Replace newlines with spaces
.replace(/\s+/g, " ") // Replace multiple spaces with a single space
.trim();
// If text length is within the allowed range, return it directly
if (plainText.length <= maxLength) {
return plainText;
}
// Take the beginning and add ellipsis
return plainText.substring(0, maxLength - 3) + "...";
}
/**
* Generate task completion summary
* @param taskName Task name
* @param taskDescription Task description
* @param completionDetails Completion details (optional)
* @returns Generated task summary
*/
export function generateTaskSummary(
taskName: string,
taskDescription: string,
completionDetails?: string
): string {
// If completion details are provided, use them preferentially
if (completionDetails) {
return extractSummary(completionDetails, 250);
}
// Otherwise generate a summary from the task name and description
const baseText = `${taskName} has been successfully completed. This task involved ${extractSummary(
taskDescription,
200
)}`;
return extractSummary(baseText, 250);
}
/**
* Split text into sentences
*
* @param text Text to split
* @returns Array of sentences
*/
function splitIntoSentences(text: string): string[] {
// Use regular expressions to split sentences
// Match periods, question marks, exclamation marks, and line breaks in both Chinese and English
const sentenceSplitters = /(?<=[。.！!？?\n])\s*/g;
const sentences = text
.split(sentenceSplitters)
.filter((s) => s.trim().length > 0);
return sentences;
}
/**
* Score sentences to determine their importance in the summary
*
* @param sentence Sentence to score
* @param index Position index of the sentence in the original text
* @param totalSentences Total number of sentences in the original text
* @returns Importance score of the sentence
*/
function scoreSentence(
sentence: string,
index: number,
totalSentences: number
): number {
let score = 1.0;
// Position factor: Sentences at the beginning and end of the document are usually more important
if (index === 0 || index === totalSentences - 1) {
score *= 1.5;
} else if (
index < Math.ceil(totalSentences * 0.2) ||
index >= Math.floor(totalSentences * 0.8)
) {
score *= 1.25;
}
// Sentence length factor: Short sentences may contain less information, long sentences may contain too much information
const wordCount = sentence.split(/\s+/).length;
if (wordCount < 3) {
score *= 0.8;
} else if (wordCount > 25) {
score *= 0.9;
} else if (wordCount >= 5 && wordCount <= 15) {
score *= 1.2;
}
// Keyword factor: Sentences containing keywords are more important
for (const [keyword, weight] of Object.entries(KEYWORDS)) {
if (sentence.includes(keyword)) {
score *= weight;
}
}
// Sentence structure factor: Special sentence patterns may be more important
if (
sentence.includes("summary") ||
sentence.includes("conclusion") ||
sentence.includes("therefore") ||
sentence.includes("thus")
) {
score *= 1.5;
}
// Numbers and proper nouns factor: Sentences containing numbers and proper nouns are usually more important
if (/\d+/.test(sentence)) {
score *= 1.3;
}
return score;
}
/**
* Extract a brief title from specified content
*
* @param content Content from which to extract the title
* @param maxLength Maximum length of the title
* @returns Extracted title
*/
export function extractTitle(content: string, maxLength: number = 50): string {
// Defensive check
if (!content || content.trim().length === 0) {
return "";
}
// Split into sentences
const sentences = splitIntoSentences(content);
if (sentences.length === 0) {
return "";
}
// First consider the first sentence
let title = sentences[0];
// If the first sentence is too long, find the first comma or other delimiter to truncate
if (title.length > maxLength) {
const firstPart = title.split(/[,，:：]/)[0];
if (firstPart && firstPart.length < maxLength) {
title = firstPart;
} else {
title = title.substring(0, maxLength - 3) + "...";
}
}
return title;
}
/**
* Intelligent extract summary based on conversation context
*
* @param messages List of conversation messages, each containing role and content
* @param maxLength Maximum length of the summary
* @returns Extracted summary
*/
export function extractSummaryFromConversation(
messages: Array<{ role: string; content: string }>,
maxLength: number = 200
): string {
// Defensive check
if (!messages || messages.length === 0) {
return "";
}
// If there's only one message, directly extract its summary
if (messages.length === 1) {
return extractSummary(messages[0].content, maxLength);
}
// Connect all messages, but retain role information
const combinedText = messages
.map((msg) => `${msg.role}: ${msg.content}`)
.join("\n");
// Extract summary from combined text
const summary = extractSummary(combinedText, maxLength);
return summary;
}