smart_rag 0.1.0

data/lib/smart_rag/services/summarization_service.rb

@@ -0,0 +1,322 @@
+require "smart_prompt"
+require_relative "../errors"
+
+module SmartRAG
+  module Services
+    # Service for generating natural language summaries and responses
+    class SummarizationService
+      attr_reader :config, :logger, :smart_prompt_engine
+
+      # Initialize the summarization service
+      # @param config [Hash] Configuration options
+      # @option config [String] :config_path Path to smart_prompt config (default: config/llm_config.yml)
+      # @option config [Integer] :max_retries Maximum retries for API calls (default: 3)
+      # @option config [Integer] :timeout Timeout for API calls (default: 30)
+      # @option config [Logger] :logger Logger instance (default: Logger.new(STDOUT))
+      # @option config [Integer] :max_context_length Maximum context length (default: 4000)
+      def initialize(config = {})
+        config ||= {}
+        @logger = Logger.new(STDOUT)
+        @config = default_config.merge(config)
+        @logger = @config[:logger] || @logger
+        @max_context_length = @config[:max_context_length]
+
+        # Load workers
+        workers_dir = File.join(File.dirname(__FILE__), '..', '..', '..', 'workers')
+        Dir.glob(File.join(workers_dir, '*.rb')).each { |file| require file }
+
+        # Initialize SmartPrompt engine
+        config_path = @config[:config_path] || "config/llm_config.yml"
+        @smart_prompt_engine = SmartPrompt::Engine.new(config_path)
+
+        @logger.info "SummarizationService initialized"
+      rescue StandardError => e
+        log_error("Failed to initialize SummarizationService", e)
+        raise
+      end
+
+      # Summarize search results into a coherent answer
+      # @param question [String] The original question
+      # @param context [String] Search results context
+      # @param options [Hash] Summarization options
+      # @option options [Symbol] :language Output language (:zh_cn, :zh_tw, :en, :ja)
+      # @option options [Integer] :max_length Maximum response length (default: 1000)
+      # @option options [String] :tone Response tone (formal, casual, technical)
+      # @option options [Boolean] :include_citations Whether to include citations (default: true)
+      # @return [Hash] Summarized response with answer and metadata
+      def summarize_search_results(question, context, options = {})
+        raise ArgumentError, "Question cannot be nil or empty" if question.to_s.strip.empty?
+        raise ArgumentError, "Context cannot be nil" if context.nil?
+
+        logger.info "Summarizing search results for question: #{question[0..50]}..."
+        logger.info "Context length: #{context.length} chars"
+
+        # Truncate context if too long
+        truncated_context = truncate_context(context)
+        language = options[:language] || :en
+        max_length = options[:max_length] || 1000
+        tone = options[:tone] || 'formal'
+
+        # Build prompt based on language
+        prompt = build_summarization_prompt(
+          question,
+          truncated_context,
+          language,
+          max_length,
+          tone,
+          options
+        )
+
+        # Call LLM to generate summary
+        response = call_llm_for_summary(prompt, options)
+
+        # Parse response
+        parsed_response = parse_summary_response(response, options)
+
+        logger.info "Successfully generated summary (#{parsed_response[:answer].length} chars), confidence: #{parsed_response[:confidence]}"
+
+        parsed_response
+      rescue ArgumentError
+        raise
+      rescue StandardError => e
+        log_error("Failed to summarize search results", e)
+        raise ::SmartRAG::Errors::SummarizationServiceError, "Summarization failed: #{e.message}"
+      end
+
+      # Generate a standalone summary of a text
+      # @param text [String] Text to summarize
+      # @param options [Hash] Summarization options
+      # @return [String] Summary text
+      def summarize_text(text, options = {})
+        raise ArgumentError, "Text cannot be nil or empty" if text.to_s.strip.empty?
+
+        logger.info "Summarizing text (#{text.length} chars)..."
+
+        language = options[:language] || detect_language(text)
+        max_length = options[:max_length] || 500
+
+        prompt = build_standalone_summary_prompt(text, language, max_length)
+        response = call_llm_for_summary(prompt, options)
+
+        summary = extract_text_from_response(response)
+        logger.info "Generated summary (#{summary.length} chars)"
+
+        summary
+      rescue ArgumentError
+        raise
+      rescue StandardError => e
+        log_error("Failed to summarize text", e)
+        raise ::SmartRAG::Errors::SummarizationServiceError, "Text summarization failed: #{e.message}"
+      end
+
+      private
+
+      def build_summarization_prompt(question, context, language, max_length, tone, options)
+        include_citations = options.fetch(:include_citations, true)
+
+        case language
+        when :zh_cn
+          build_chinese_summarization_prompt(question, context, max_length, tone, include_citations)
+        when :zh_tw
+          build_traditional_chinese_summarization_prompt(question, context, max_length, tone, include_citations)
+        when :en
+          build_english_summarization_prompt(question, context, max_length, tone, include_citations)
+        when :ja
+          build_japanese_summarization_prompt(question, context, max_length, tone, include_citations)
+        else
+          logger.warn "Unsupported language: #{language}, defaulting to English"
+          build_english_summarization_prompt(question, context, max_length, tone, include_citations)
+        end
+      end
+
+      def build_chinese_summarization_prompt(question, context, max_length, tone, include_citations)
+        prompt = "基于以下搜索结果，回答问题并提供详细解释。"
+        prompt << "\n\n问题：#{question}\n\n"
+        prompt << "搜索结果：\n#{context}\n\n"
+        prompt << "要求：\n"
+        prompt << "1. 提供直接、准确的答案\n"
+        prompt << "2. 使用搜索结果中的信息支持你的回答\n"
+        prompt << "3. 答案长度不超过#{max_length}个字符\n"
+        prompt << "4. 语气：#{tone == 'formal' ? '正式' : tone == 'casual' ? '随意' : '专业'}\n"
+        prompt << "5. #{include_citations ? '使用[1]、[2]等格式引用来源' : '不需要引用来源'}\n\n"
+        prompt << "请提供结构化的回答：\n"
+        prompt << "- 简要答案（1-2句话）\n"
+        prompt << "- 详细解释\n"
+        include_citations ? (prompt << "- 来源引用\n") : ""
+        prompt << "\n以JSON格式输出：{\"answer\": \"...\", \"confidence\": 0.0-1.0}"
+      end
+
+      def build_traditional_chinese_summarization_prompt(question, context, max_length, tone, include_citations)
+        prompt = "基於以下搜尋結果，回答問題並提供詳細解釋。"
+        prompt << "\n\n問題：#{question}\n\n"
+        prompt << "搜尋結果：\n#{context}\n\n"
+        prompt << "要求：\n"
+        prompt << "1. 提供直接、準確的答案\n"
+        prompt << "2. 使用搜尋結果中的資訊支持你的回答\n"
+        prompt << "3. 答案長度不超過#{max_length}個字元\n"
+        prompt << "4. 語氣：#{tone == 'formal' ? '正式' : tone == 'casual' ? '隨意' : '專業'}\n"
+        prompt << "5. #{include_citations ? '使用[1]、[2]等格式引用來源' : '不需要引用來源'}\n\n"
+        prompt << "請提供結構化的回答：\n"
+        prompt << "- 簡要答案（1-2句話）\n"
+        prompt << "- 詳細解釋\n"
+        include_citations ? (prompt << "- 來源引用\n") : ""
+        prompt << "\n以JSON格式輸出：{\"answer\": \"...\", \"confidence\": 0.0-1.0}"
+      end
+
+      def build_english_summarization_prompt(question, context, max_length, tone, include_citations)
+        prompt = "Based on the following search results, answer the question and provide detailed explanation."
+        prompt << "\n\nQuestion: #{question}\n\n"
+        prompt << "Search Results:\n#{context}\n\n"
+        prompt << "Requirements:\n"
+        prompt << "1. Provide a direct, accurate answer\n"
+        prompt << "2. Support your answer with information from the search results\n"
+        prompt << "3. Keep answer under #{max_length} characters\n"
+        prompt << "4. Tone: #{tone}\n"
+        prompt << "5. #{include_citations ? 'Cite sources using [1], [2] format' : 'No citations needed'}\n\n"
+        prompt << "Provide a structured response:\n"
+        prompt << "- Brief answer (1-2 sentences)\n"
+        prompt << "- Detailed explanation\n"
+        include_citations ? (prompt << "- Source citations\n") : ""
+        prompt << "\nOutput in JSON format: {\"answer\": \"...\", \"confidence\": 0.0-1.0}"
+      end
+
+      def build_japanese_summarization_prompt(question, context, max_length, tone, include_citations)
+        prompt = "以下の検索結果に基づいて、質問に答えて詳細な説明を提供してください。"
+        prompt << "\n\n質問：#{question}\n\n"
+        prompt << "検索結果：\n#{context}\n\n"
+        prompt << "要件：\n"
+        prompt << "1. 直接的で正確な答えを提供する\n"
+        prompt << "2. 検索結果の情報を使用して回答をサポートする\n"
+        prompt << "3. 回答は#{max_length}文字以内にする\n"
+        prompt << "4. トーン：#{tone == 'formal' ? 'フォーマル' : tone == 'casual' ? 'カジュアル' : '専門的'}\n"
+        prompt << "5. #{include_citations ? '[1]、[2]などの形式で情報源を引用' : '引用は不要'}\n\n"
+        prompt << "構造化された回答を提供：\n"
+        prompt << "- 簡潔な答え（1-2文）\n"
+        prompt << "- 詳細な説明\n"
+        include_citations ? (prompt << "- 情報源の引用\n") : ""
+        prompt << "\nJSON形式で出力：{\"answer\": \"...\", \"confidence\": 0.0-1.0}"
+      end
+
+      def build_standalone_summary_prompt(text, language, max_length)
+        case language
+        when :zh_cn
+          "用#{max_length}字以内的简洁中文总结以下内容：\n\n#{text}"
+        when :zh_tw
+          "用#{max_length}字以內的簡潔繁體中文總結以下內容：\n\n#{text}"
+        when :en
+          "Summarize the following content in English within #{max_length} characters:\n\n#{text}"
+        when :ja
+          "次の内容を日本語で#{max_length}文字以内に要約してください：\n\n#{text}"
+        else
+          "Summarize the following content within #{max_length} characters:\n\n#{text}"
+        end
+      end
+
+      def call_llm_for_summary(prompt, options = {})
+        max_retries = options[:retries] || config[:max_retries]
+        timeout = options[:timeout] || config[:timeout]
+
+        with_retry(max_retries: max_retries, timeout: timeout) do
+          result = smart_prompt_engine.call_worker(:generate_content, { content: prompt })
+          raise "No response from LLM" unless result
+          result
+        end
+      rescue StandardError => e
+        logger.error "LLM call for summarization failed: #{e.message}"
+        raise
+      end
+
+      def parse_summary_response(response, options = {})
+        # Try to parse as JSON first
+        if response =~ /\{.*answer.*confidence.*\}/m
+          begin
+            parsed = JSON.parse(response.gsub(/```json\n?|\n?```/, ''))
+            return {
+              answer: parsed["answer"] || parsed["response"] || response,
+              confidence: parsed["confidence"]&.to_f || 0.8,
+              raw_response: response
+            }
+          rescue JSON::ParserError
+            logger.warn "Failed to parse JSON response, using raw response"
+          end
+        end
+
+        # Fallback to using the entire response as answer
+        {
+          answer: response,
+          confidence: 0.8, # Default confidence
+          raw_response: response
+        }
+      end
+
+      def extract_text_from_response(response)
+        # Remove any JSON wrapper if present
+        if response =~ /\{.*answer.*\}/m
+          begin
+            parsed = JSON.parse(response.gsub(/```json\n?|\n?```/, ''))
+            return parsed["answer"] || parsed["response"] || response
+          rescue JSON::ParserError
+            # Continue to fallback
+          end
+        end
+
+        # Remove markdown code blocks
+        response.gsub(/```[a-z]*\n?|\n?```/, '').strip
+      end
+
+      def truncate_context(context)
+        return context if context.length <= max_context_length
+
+        logger.warn "Context too long (#{context.length} chars), truncating to #{max_context_length}"
+        context[0...max_context_length] + "... (truncated)"
+      end
+
+      def detect_language(text)
+        # Check for Japanese hiragana/katakana first (more specific than Chinese kanji)
+        return :ja if text.match?(/[\u3040-\u309f\u30a0-\u30ff]/)
+        return :zh_cn if text.match?(/[\u4e00-\u9fff]/)
+        :en
+      end
+
+      def max_context_length
+        @max_context_length
+      end
+
+      def with_retry(max_retries:, timeout:, &block)
+        last_exception = nil
+
+        max_retries.times do |attempt|
+          begin
+            Timeout.timeout(timeout) do
+              return yield
+            end
+          rescue StandardError => e
+            last_exception = e
+            logger.warn "Attempt #{attempt + 1} failed: #{e.message}"
+
+            # Exponential backoff
+            sleep(2  ** attempt) if attempt < max_retries - 1
+          end
+        end
+
+        raise last_exception
+      end
+
+      def log_error(message, exception)
+        active_logger = logger || @logger || Logger.new(STDOUT)
+        active_logger.error "#{message}: #{exception.message}"
+        active_logger.error exception.backtrace.join("\n  ")
+      end
+
+      def default_config
+        {
+          config_path: "config/llm_config.yml",
+          max_retries: 3,
+          timeout: 30,
+          max_context_length: 4000,
+          logger: Logger.new(STDOUT)
+        }
+      end
+    end
+  end
+end