src/lib/local-summarizer.ts

/**
 * @fileoverview A more intelligent, local, rule-based algorithm for summarizing chats.
 * This acts as a "primitive AI" for summarization, running entirely in the browser.
 */

import type { Message, User } from './types';

export interface LocalSummary {
    participants: string[];
    topic: string;
    summaryPoints: string[];
}

// Simple list of common "stop words" to ignore when finding topics.
const STOP_WORDS_EN = new Set(['a', 'an', 'the', 'is', 'are', 'was', 'were', 'in', 'on', 'at', 'i', 'you', 'he', 'she', 'it', 'we', 'they', 'and', 'but', 'or', 'so', 'to', 'of', 'for', 'with', 'about', 'as', 'that', 'this', 'what', 'when', 'where', 'why', 'how', 'do', 'does', 'did']);
const STOP_WORDS_AR = new Set(['أنا', 'أنت', 'هو', 'هي', 'نحن', 'هم', 'و', 'أو', 'في', 'على', 'من', 'عن', 'ما', 'ماذا', 'كيف', 'متى', 'أين', 'هل', 'كان', 'يكون', 'قال', 'قالت', 'هذا', 'هذه', 'ذلك', 'تلك']);

// Keywords that indicate an important message
const KEYWORD_SCORES = {
    // English
    'agree': 5, 'decision': 5, 'plan': 5, 'important': 5, 'because': 4, 'question': 4,
    'next step': 5, 'finally': 4, 'conclusion': 5, 'i will': 3, 'we should': 3,
    // Arabic
    'موافقة': 5, 'قرار': 5, 'خطة': 5, 'مهم': 5, 'لأن': 4, 'سؤال': 4,
    'الخطوة التالية': 5, 'أخيرا': 4, 'خلاصة': 5, 'سأقوم': 3, 'يجب علينا': 3, 'اتفقنا': 5,
};

interface ScoredMessage {
    message: Message;
    score: number;
}

/**
 * Scores a message based on its content to determine its importance.
 * @param message The message to score.
 * @param previousMessage The message that came before it.
 * @returns A numerical score.
 */
function getMessageScore(message: Message, previousMessage?: Message): number {
    if (!message.text || message.isSystemMessage) return 0;

    let score = 0;
    const text = message.text.toLowerCase();

    // 1. Keyword scoring
    for (const keyword in KEYWORD_SCORES) {
        if (text.includes(keyword)) {
            score += KEYWORD_SCORES[keyword as keyof typeof KEYWORD_SCORES];
        }
    }

    // 2. Question scoring
    if (text.includes('?')) {
        score += 5; // Questions are very important
    }

    // 3. Answer scoring (if it follows a question)
    if (previousMessage && previousMessage.text?.includes('?')) {
        score += 6; // Answers are even more important
    }

    // 4. Length scoring (longer messages are likely more detailed)
    if (text.length > 100) {
        score += 3;
    } else if (text.length > 50) {
        score += 2;
    }

    return score;
}

/**
 * Generates a more intelligent summary of a chat history locally.
 * @param messages The array of messages in the chat.
 * @param currentUser The currently logged-in user.
 * @returns A LocalSummary object.
 */
export function generateLocalSummary(messages: Message[], currentUser: User): LocalSummary {
    if (messages.length === 0) {
        return {
            participants: [],
            topic: 'No conversation yet',
            summaryPoints: ['The chat is empty.'],
        };
    }

    // 1. Identify Participants
    const participantSet = new Set<string>();
    messages.forEach(msg => {
        if (msg.sender === currentUser.uid) {
            participantSet.add(currentUser.displayName);
        } else {
            participantSet.add(msg.senderDisplayName);
        }
    });
    const participants = Array.from(participantSet);

    // 2. Score all messages for importance
    const scoredMessages: ScoredMessage[] = [];
    for (let i = 0; i < messages.length; i++) {
        const message = messages[i];
        const previousMessage = i > 0 ? messages[i - 1] : undefined;
        scoredMessages.push({
            message,
            score: getMessageScore(message, previousMessage),
        });
    }

    // Sort messages by score to find the most important ones
    const sortedScoredMessages = scoredMessages.sort((a, b) => b.score - a.score);

    // 3. Determine Topic from the most important messages
    const wordCounts = new Map<string, number>();
    const stopWords = new Set([...STOP_WORDS_EN, ...STOP_WORDS_AR]);
    
    // Analyze top 5 most important messages for topic words
    sortedScoredMessages.slice(0, 5).forEach(({ message }) => {
        if (!message.text) return;
        const words = message.text.toLowerCase().split(/\s+/);
        words.forEach(word => {
            const cleanWord = word.replace(/[.,!?"'()؟]/g, '');
            if (cleanWord.length > 3 && !stopWords.has(cleanWord)) {
                wordCounts.set(cleanWord, (wordCounts.get(cleanWord) || 0) + 1);
            }
        });
    });

    const sortedWords = Array.from(wordCounts.entries()).sort((a, b) => b[1] - a[1]);
    const topic = sortedWords.length > 0 ? sortedWords.slice(0, 2).map(entry => entry[0]).join(' & ') : 'General discussion';

    // 4. Generate Summary Points from the top scored messages
    const summaryPoints: string[] = [];
    const usedMessageIds = new Set<string>();

    // Take top 3-4 most important messages to form the summary
    for (const scoredMsg of sortedScoredMessages) {
        if (summaryPoints.length >= 4) break;
        if (scoredMsg.score > 0 && scoredMsg.message.id && !usedMessageIds.has(scoredMsg.message.id)) {
            const point = `${scoredMsg.message.senderDisplayName} discussed: "${scoredMsg.message.text!.substring(0, 50)}...".`;
            summaryPoints.push(point);
            usedMessageIds.add(scoredMsg.message.id);
        }
    }

    // Fallback if no "important" messages were found
    if (summaryPoints.length === 0 && messages.length > 0) {
        const firstMessage = messages[0];
        const lastMessage = messages[messages.length - 1];
        summaryPoints.push(`Conversation started by ${firstMessage.senderDisplayName}.`);
        if (lastMessage.id !== firstMessage.id) {
            summaryPoints.push(`The last message was from ${lastMessage.senderDisplayName}.`);
        }
    }

    return {
        participants,
        topic,
        summaryPoints,
    };
}