312 lines
9.7 KiB
JavaScript
312 lines
9.7 KiB
JavaScript
// js/chatbot/core/smart-granularity-selector.js
|
||
// 智能粒度选择器 - 根据问题类型和意群特征自动选择最佳粒度
|
||
(function(window) {
|
||
'use strict';
|
||
|
||
/**
|
||
* 问题类型分类
|
||
* - overview: 概览性问题（需要summary）
|
||
* - specific: 具体细节问题（需要digest或full）
|
||
* - extraction: 信息提取问题（需要full）
|
||
* - analytical: 分析性问题（需要digest）
|
||
*/
|
||
const QUERY_PATTERNS = {
|
||
overview: [
|
||
/总结|概括|概述|简述|大意|主要内容|主题|讲.*什么|关于什么/,
|
||
/整体|全文|全部|所有|overall|summary|general/i,
|
||
/介绍|背景|目的|意义|作用/,
|
||
/有哪些|包括.*什么|涉及.*什么/
|
||
],
|
||
extraction: [
|
||
/具体|详细|准确|精确|原文|exact|specific|detail/i,
|
||
/数据|数值|数字|结果|table|figure|chart/i,
|
||
/步骤|流程|过程|方法|algorithm|procedure/i,
|
||
/公式|方程|equation|formula/,
|
||
/引用|citation|reference|出处/,
|
||
/代码|code|实现|implementation/
|
||
],
|
||
analytical: [
|
||
/分析|解释|说明|explain|analyze|why|how/i,
|
||
/原因|理由|依据|根据|原理|机制/,
|
||
/比较|对比|区别|差异|联系|关系|compare/i,
|
||
/优缺点|利弊|advantage|disadvantage/,
|
||
/影响|作用|效果|impact|effect/
|
||
]
|
||
};
|
||
|
||
/**
|
||
* 粒度选择规则
|
||
* - summary: 摘要（80字）- 用于快速浏览、索引匹配
|
||
* - digest: 精要（1000字）- 用于一般性分析、问答
|
||
* - full: 全文（完整文本）- 用于精确查找、详细分析
|
||
*/
|
||
const GRANULARITY_RULES = {
|
||
overview: {
|
||
default: 'summary',
|
||
maxGroups: 10, // 概览问题可以返回更多意群
|
||
description: '概览性查询：使用摘要快速扫描'
|
||
},
|
||
analytical: {
|
||
default: 'digest',
|
||
maxGroups: 5,
|
||
description: '分析性查询：使用精要提供足够细节'
|
||
},
|
||
extraction: {
|
||
default: 'full',
|
||
maxGroups: 3, // 精确查询限制意群数量，避免上下文过长
|
||
description: '提取性查询：使用全文确保信息完整'
|
||
},
|
||
specific: {
|
||
default: 'digest',
|
||
maxGroups: 5,
|
||
description: '具体性查询：使用精要平衡细节与长度'
|
||
}
|
||
};
|
||
|
||
/**
|
||
* 分析查询类型
|
||
* @param {string} query - 用户查询
|
||
* @returns {string} 查询类型 (overview/analytical/extraction/specific)
|
||
*/
|
||
function analyzeQueryType(query) {
|
||
const q = String(query || '').trim();
|
||
if (!q) return 'specific';
|
||
|
||
// 检查各类型模式
|
||
for (const [type, patterns] of Object.entries(QUERY_PATTERNS)) {
|
||
for (const pattern of patterns) {
|
||
if (pattern.test(q)) {
|
||
return type;
|
||
}
|
||
}
|
||
}
|
||
|
||
// 默认使用specific类型
|
||
return 'specific';
|
||
}
|
||
|
||
/**
|
||
* 根据意群特征调整粒度
|
||
* @param {Object} group - 意群对象
|
||
* @param {string} baseGranularity - 基础粒度
|
||
* @returns {string} 调整后的粒度
|
||
*/
|
||
function adjustByGroupFeatures(group, baseGranularity) {
|
||
if (!group) return baseGranularity;
|
||
|
||
const charCount = group.charCount || 0;
|
||
const hasDigest = !!(group.digest && group.digest.length > 100);
|
||
const hasFull = !!(group.fullText && group.fullText.length > 500);
|
||
|
||
// 如果意群本身很短（<2000字），直接使用full
|
||
if (charCount < 2000 && hasFull) {
|
||
return 'full';
|
||
}
|
||
|
||
// 如果没有digest，降级到summary或升级到full
|
||
if (!hasDigest) {
|
||
if (baseGranularity === 'digest') {
|
||
return hasFull ? 'full' : 'summary';
|
||
}
|
||
}
|
||
|
||
// 如果没有full，digest是最高粒度
|
||
if (!hasFull && baseGranularity === 'full') {
|
||
return hasDigest ? 'digest' : 'summary';
|
||
}
|
||
|
||
return baseGranularity;
|
||
}
|
||
|
||
/**
|
||
* 智能选择粒度
|
||
* @param {string} query - 用户查询
|
||
* @param {Array} groups - 候选意群列表
|
||
* @param {Object} options - 选项
|
||
* @returns {Object} { granularity, maxGroups, queryType, reasoning }
|
||
*/
|
||
function selectGranularity(query, groups = [], options = {}) {
|
||
// 分析查询类型
|
||
const queryType = analyzeQueryType(query);
|
||
const rule = GRANULARITY_RULES[queryType] || GRANULARITY_RULES.specific;
|
||
|
||
let granularity = options.forceGranularity || rule.default;
|
||
let maxGroups = options.maxGroups || rule.maxGroups;
|
||
|
||
// 如果候选意群少，可以使用更高粒度
|
||
if (groups.length <= 2 && granularity === 'summary') {
|
||
granularity = 'digest';
|
||
} else if (groups.length === 1 && granularity !== 'full') {
|
||
granularity = 'full';
|
||
}
|
||
|
||
// Token限制检查
|
||
const estimatedTokens = estimateTokenUsage(groups.slice(0, maxGroups), granularity);
|
||
if (options.maxTokens && estimatedTokens > options.maxTokens) {
|
||
// Token超限，降级粒度或减少意群数
|
||
if (granularity === 'full') {
|
||
granularity = 'digest';
|
||
} else if (granularity === 'digest' && estimatedTokens > options.maxTokens * 1.5) {
|
||
granularity = 'summary';
|
||
} else {
|
||
// 减少意群数量
|
||
maxGroups = Math.max(1, Math.floor(maxGroups * 0.6));
|
||
}
|
||
}
|
||
|
||
return {
|
||
granularity,
|
||
maxGroups,
|
||
queryType,
|
||
reasoning: rule.description,
|
||
estimatedTokens: estimateTokenUsage(groups.slice(0, maxGroups), granularity)
|
||
};
|
||
}
|
||
|
||
/**
|
||
* 估算Token使用量
|
||
* @param {Array} groups - 意群列表
|
||
* @param {string} granularity - 粒度
|
||
* @returns {number} 估算的token数
|
||
*/
|
||
function estimateTokenUsage(groups, granularity) {
|
||
if (!Array.isArray(groups) || groups.length === 0) return 0;
|
||
|
||
let totalChars = 0;
|
||
groups.forEach(g => {
|
||
if (granularity === 'summary') {
|
||
totalChars += (g.summary || '').length;
|
||
} else if (granularity === 'digest') {
|
||
totalChars += (g.digest || '').length;
|
||
} else if (granularity === 'full') {
|
||
totalChars += g.charCount || (g.fullText || '').length;
|
||
}
|
||
});
|
||
|
||
// 中文平均1.5字符=1token，英文平均4字符=1token
|
||
// 这里简化为平均2字符=1token
|
||
return Math.ceil(totalChars / 2);
|
||
}
|
||
|
||
/**
|
||
* 批量选择意群的粒度（支持混合粒度）
|
||
* @param {string} query - 用户查询
|
||
* @param {Array} rankedGroups - 已排序的候选意群（相关性从高到低）
|
||
* @param {Object} options - 选项
|
||
* @returns {Array} [ { group, granularity, score } ]
|
||
*/
|
||
function selectMixedGranularity(query, rankedGroups = [], options = {}) {
|
||
const queryType = analyzeQueryType(query);
|
||
const baseRule = GRANULARITY_RULES[queryType] || GRANULARITY_RULES.specific;
|
||
|
||
const maxTokens = options.maxTokens || 8000;
|
||
const result = [];
|
||
let accumulatedTokens = 0;
|
||
|
||
for (let i = 0; i < rankedGroups.length; i++) {
|
||
const item = rankedGroups[i];
|
||
const group = item.group || item;
|
||
const score = item.score || 1.0;
|
||
|
||
// 排名越靠前，使用越高粒度
|
||
let granularity;
|
||
if (i === 0) {
|
||
// 最相关的意群：使用最高粒度
|
||
granularity = queryType === 'overview' ? 'digest' : 'full';
|
||
} else if (i < 3) {
|
||
// 前3个意群：使用中等粒度
|
||
granularity = baseRule.default;
|
||
} else {
|
||
// 其他意群：使用低粒度
|
||
granularity = 'summary';
|
||
}
|
||
|
||
// 根据意群特征调整
|
||
granularity = adjustByGroupFeatures(group, granularity);
|
||
|
||
// 估算token
|
||
const tokens = estimateTokenUsage([group], granularity);
|
||
|
||
// 检查是否会超限
|
||
if (accumulatedTokens + tokens > maxTokens) {
|
||
// 尝试降级
|
||
if (granularity === 'full') {
|
||
granularity = 'digest';
|
||
} else if (granularity === 'digest') {
|
||
granularity = 'summary';
|
||
} else {
|
||
// 已经是summary，无法继续添加
|
||
break;
|
||
}
|
||
}
|
||
|
||
const finalTokens = estimateTokenUsage([group], granularity);
|
||
if (accumulatedTokens + finalTokens > maxTokens) {
|
||
break; // 无法添加更多意群
|
||
}
|
||
|
||
accumulatedTokens += finalTokens;
|
||
result.push({
|
||
group,
|
||
granularity,
|
||
score,
|
||
tokens: finalTokens
|
||
});
|
||
|
||
// 检查是否已有足够的意群
|
||
if (result.length >= (baseRule.maxGroups || 5)) {
|
||
break;
|
||
}
|
||
}
|
||
|
||
return result;
|
||
}
|
||
|
||
/**
|
||
* 构建混合粒度上下文
|
||
* @param {Array} selections - selectMixedGranularity的结果
|
||
* @returns {string} 组合后的上下文文本
|
||
*/
|
||
function buildMixedContext(selections) {
|
||
if (!Array.isArray(selections) || selections.length === 0) return '';
|
||
|
||
const parts = [];
|
||
selections.forEach(sel => {
|
||
const g = sel.group;
|
||
const gran = sel.granularity;
|
||
|
||
let text = '';
|
||
if (gran === 'summary') {
|
||
text = g.summary || '';
|
||
} else if (gran === 'digest') {
|
||
text = g.digest || g.summary || '';
|
||
} else if (gran === 'full') {
|
||
text = g.fullText || g.digest || g.summary || '';
|
||
}
|
||
|
||
if (text) {
|
||
const keywords = (g.keywords || []).join('、');
|
||
parts.push(`【${g.groupId} - ${gran}】\n关键词: ${keywords}\n内容:\n${text}`);
|
||
}
|
||
});
|
||
|
||
return parts.join('\n\n');
|
||
}
|
||
|
||
// 导出
|
||
window.SmartGranularitySelector = {
|
||
analyzeQueryType,
|
||
selectGranularity,
|
||
selectMixedGranularity,
|
||
buildMixedContext,
|
||
estimateTokenUsage,
|
||
adjustByGroupFeatures,
|
||
// 暴露规则以便测试和调整
|
||
GRANULARITY_RULES,
|
||
QUERY_PATTERNS
|
||
};
|
||
|
||
console.log('[SmartGranularitySelector] 智能粒度选择器已加载');
|
||
|
||
})(window);
|