dramaling-vocab-learning/backend/DramaLing.Api/Services/AI/GeminiAIProvider.cs

482 lines
16 KiB
C#

using DramaLing.Api.Models.DTOs;
using DramaLing.Api.Models.Configuration;
using Microsoft.Extensions.Options;
using System.Text.Json;
using System.Text;
using System.Diagnostics;
namespace DramaLing.Api.Services.AI;
/// <summary>
/// Google Gemini AI 提供商實作
/// </summary>
public class GeminiAIProvider : IAIProvider
{
private readonly HttpClient _httpClient;
private readonly ILogger<GeminiAIProvider> _logger;
private readonly GeminiOptions _options;
private AIProviderStats _stats;
public GeminiAIProvider(HttpClient httpClient, IOptions<GeminiOptions> options, ILogger<GeminiAIProvider> logger)
{
_httpClient = httpClient ?? throw new ArgumentNullException(nameof(httpClient));
_logger = logger ?? throw new ArgumentNullException(nameof(logger));
_options = options.Value ?? throw new ArgumentNullException(nameof(options));
_stats = new AIProviderStats();
_httpClient.Timeout = TimeSpan.FromSeconds(_options.TimeoutSeconds);
_httpClient.DefaultRequestHeaders.Add("User-Agent", "DramaLing/1.0");
_logger.LogInformation("GeminiAIProvider initialized with model: {Model}, timeout: {Timeout}s",
_options.Model, _options.TimeoutSeconds);
}
#region IAIProvider
public string ProviderName => "Google Gemini";
public bool IsAvailable => !string.IsNullOrEmpty(_options.ApiKey);
public decimal CostPerRequest => 0.001m; // 大概每次請求成本
public int MaxInputLength => _options.MaxOutputTokens / 4; // 粗略估計
public int AverageResponseTimeMs => _stats.AverageResponseTimeMs;
#endregion
#region
public async Task<SentenceAnalysisData> AnalyzeSentenceAsync(string inputText, AnalysisOptions options)
{
var stopwatch = Stopwatch.StartNew();
_stats.TotalRequests++;
try
{
_logger.LogInformation("Starting sentence analysis for text: {Text}",
inputText.Substring(0, Math.Min(50, inputText.Length)));
var prompt = BuildAnalysisPrompt(inputText);
var aiResponse = await CallGeminiAPIAsync(prompt);
if (string.IsNullOrWhiteSpace(aiResponse))
{
throw new InvalidOperationException("Gemini API returned empty response");
}
var analysisData = ParseAIResponse(inputText, aiResponse);
stopwatch.Stop();
RecordSuccessfulRequest(stopwatch.ElapsedMilliseconds);
_logger.LogInformation("Sentence analysis completed in {ElapsedMs}ms", stopwatch.ElapsedMilliseconds);
return analysisData;
}
catch (Exception ex)
{
stopwatch.Stop();
RecordFailedRequest(stopwatch.ElapsedMilliseconds);
_logger.LogError(ex, "Error analyzing sentence: {Text}", inputText);
throw;
}
}
public async Task<AIProviderHealthStatus> CheckHealthAsync()
{
var stopwatch = Stopwatch.StartNew();
try
{
var testPrompt = "Test health check prompt";
var response = await CallGeminiAPIAsync(testPrompt);
stopwatch.Stop();
return new AIProviderHealthStatus
{
IsHealthy = !string.IsNullOrEmpty(response),
CheckedAt = DateTime.UtcNow,
ResponseTimeMs = (int)stopwatch.ElapsedMilliseconds
};
}
catch (Exception ex)
{
stopwatch.Stop();
return new AIProviderHealthStatus
{
IsHealthy = false,
ErrorMessage = ex.Message,
CheckedAt = DateTime.UtcNow,
ResponseTimeMs = (int)stopwatch.ElapsedMilliseconds
};
}
}
public Task<AIProviderStats> GetStatsAsync()
{
return Task.FromResult(_stats);
}
#endregion
#region
private string BuildAnalysisPrompt(string inputText)
{
return $@"You are an English learning assistant. Analyze this sentence and return ONLY a valid JSON response.
**Input Sentence**: ""{inputText}""
**Required JSON Structure:**
{{
""sentenceTranslation"": ""Traditional Chinese translation of the entire sentence"",
""hasGrammarErrors"": true/false,
""grammarCorrections"": [
{{
""original"": ""incorrect text"",
""corrected"": ""correct text"",
""type"": ""error type (tense/subject-verb/preposition/word-order)"",
""explanation"": ""brief explanation in Traditional Chinese""
}}
],
""vocabularyAnalysis"": {{
""word1"": {{
""word"": ""the word"",
""translation"": ""Traditional Chinese translation"",
""definition"": ""English definition"",
""partOfSpeech"": ""noun/verb/adjective/etc"",
""pronunciation"": ""/phonetic/"",
""difficultyLevel"": ""A1/A2/B1/B2/C1/C2"",
""frequency"": ""high/medium/low"",
""synonyms"": [""synonym1"", ""synonym2""],
""example"": ""example sentence"",
""exampleTranslation"": ""Traditional Chinese example translation""
}}
}},
""idioms"": [
{{
""idiom"": ""idiomatic expression"",
""translation"": ""Traditional Chinese meaning"",
""definition"": ""English explanation"",
""pronunciation"": ""/phonetic notation/"",
""difficultyLevel"": ""A1/A2/B1/B2/C1/C2"",
""frequency"": ""high/medium/low"",
""synonyms"": [""synonym1"", ""synonym2""],
""example"": ""usage example"",
""exampleTranslation"": ""Traditional Chinese example""
}}
]
}}
**Analysis Guidelines:**
1. **Grammar Check**: Detect tense errors, subject-verb agreement, preposition usage, word order
2. **Vocabulary Analysis**: Include ALL significant words (exclude articles: a, an, the)
3. **CEFR Levels**: Assign accurate A1-C2 levels for each word
4. **Idioms**: Identify any idiomatic expressions or phrasal verbs
5. **Translations**: Use Traditional Chinese (Taiwan standard)
**IMPORTANT**: Return ONLY the JSON object, no additional text or explanation.";
}
private async Task<string> CallGeminiAPIAsync(string prompt)
{
try
{
var requestBody = new
{
contents = new[]
{
new
{
parts = new[]
{
new { text = prompt }
}
}
},
generationConfig = new
{
temperature = _options.Temperature,
topK = 40,
topP = 0.95,
maxOutputTokens = _options.MaxOutputTokens
}
};
var json = JsonSerializer.Serialize(requestBody);
var content = new StringContent(json, Encoding.UTF8, "application/json");
var response = await _httpClient.PostAsync(
$"{_options.BaseUrl}/v1beta/models/{_options.Model}:generateContent?key={_options.ApiKey}",
content);
response.EnsureSuccessStatusCode();
var responseJson = await response.Content.ReadAsStringAsync();
_logger.LogDebug("Raw Gemini API response: {Response}",
responseJson.Substring(0, Math.Min(500, responseJson.Length)));
return ExtractTextFromResponse(responseJson);
}
catch (Exception ex)
{
_logger.LogError(ex, "Gemini API call failed");
throw;
}
}
private string ExtractTextFromResponse(string responseJson)
{
using var document = JsonDocument.Parse(responseJson);
var root = document.RootElement;
if (root.TryGetProperty("candidates", out var candidatesElement) &&
candidatesElement.ValueKind == JsonValueKind.Array)
{
var firstCandidate = candidatesElement.EnumerateArray().FirstOrDefault();
if (firstCandidate.ValueKind != JsonValueKind.Undefined &&
firstCandidate.TryGetProperty("content", out var contentElement) &&
contentElement.TryGetProperty("parts", out var partsElement) &&
partsElement.ValueKind == JsonValueKind.Array)
{
var firstPart = partsElement.EnumerateArray().FirstOrDefault();
if (firstPart.TryGetProperty("text", out var textElement))
{
return textElement.GetString() ?? string.Empty;
}
}
}
// 檢查是否有安全過濾
if (root.TryGetProperty("promptFeedback", out _))
{
_logger.LogWarning("Gemini content filtered due to safety policies");
return "The content analysis is temporarily unavailable due to safety filtering.";
}
return string.Empty;
}
private SentenceAnalysisData ParseAIResponse(string inputText, string aiResponse)
{
try
{
var cleanJson = CleanAIResponse(aiResponse);
var aiAnalysis = JsonSerializer.Deserialize<AiAnalysisResponse>(cleanJson, new JsonSerializerOptions
{
PropertyNameCaseInsensitive = true
});
if (aiAnalysis == null)
{
throw new InvalidOperationException("Failed to parse AI response JSON");
}
return new SentenceAnalysisData
{
OriginalText = inputText,
SentenceMeaning = aiAnalysis.SentenceTranslation ?? "",
VocabularyAnalysis = ConvertVocabularyAnalysis(aiAnalysis.VocabularyAnalysis ?? new()),
Idioms = ConvertIdioms(aiAnalysis.Idioms ?? new()),
GrammarCorrection = ConvertGrammarCorrection(aiAnalysis),
Metadata = new AnalysisMetadata
{
ProcessingDate = DateTime.UtcNow,
AnalysisModel = _options.Model,
AnalysisVersion = "2.0"
}
};
}
catch (JsonException ex)
{
_logger.LogError(ex, "Failed to parse AI response as JSON: {Response}", aiResponse);
return CreateFallbackAnalysis(inputText, aiResponse);
}
}
private string CleanAIResponse(string aiResponse)
{
var cleanJson = aiResponse.Trim();
if (cleanJson.StartsWith("```json"))
{
cleanJson = cleanJson.Substring(7);
}
if (cleanJson.EndsWith("```"))
{
cleanJson = cleanJson.Substring(0, cleanJson.Length - 3);
}
return cleanJson.Trim();
}
private Dictionary<string, VocabularyAnalysisDto> ConvertVocabularyAnalysis(Dictionary<string, AiVocabularyAnalysis> aiVocab)
{
var result = new Dictionary<string, VocabularyAnalysisDto>();
foreach (var kvp in aiVocab)
{
var aiWord = kvp.Value;
result[kvp.Key] = new VocabularyAnalysisDto
{
Word = aiWord.Word ?? kvp.Key,
Translation = aiWord.Translation ?? "",
Definition = aiWord.Definition ?? "",
PartOfSpeech = aiWord.PartOfSpeech ?? "unknown",
Pronunciation = aiWord.Pronunciation ?? $"/{kvp.Key}/",
DifficultyLevel = aiWord.DifficultyLevel ?? "A2",
Frequency = aiWord.Frequency ?? "medium",
Synonyms = aiWord.Synonyms ?? new List<string>(),
Example = aiWord.Example,
ExampleTranslation = aiWord.ExampleTranslation,
};
}
return result;
}
private List<IdiomDto> ConvertIdioms(List<AiIdiom> aiIdioms)
{
var result = new List<IdiomDto>();
foreach (var aiIdiom in aiIdioms)
{
result.Add(new IdiomDto
{
Idiom = aiIdiom.Idiom ?? "",
Translation = aiIdiom.Translation ?? "",
Definition = aiIdiom.Definition ?? "",
Pronunciation = aiIdiom.Pronunciation ?? "",
DifficultyLevel = aiIdiom.DifficultyLevel ?? "B2",
Frequency = aiIdiom.Frequency ?? "medium",
Synonyms = aiIdiom.Synonyms ?? new List<string>(),
Example = aiIdiom.Example,
ExampleTranslation = aiIdiom.ExampleTranslation
});
}
return result;
}
private GrammarCorrectionDto? ConvertGrammarCorrection(AiAnalysisResponse aiAnalysis)
{
if (!aiAnalysis.HasGrammarErrors || aiAnalysis.GrammarCorrections == null || !aiAnalysis.GrammarCorrections.Any())
{
return null;
}
var corrections = aiAnalysis.GrammarCorrections.Select(gc => new GrammarErrorDto
{
Error = gc.Original ?? "",
Correction = gc.Corrected ?? "",
Type = gc.Type ?? "grammar",
Explanation = gc.Explanation ?? "",
Severity = "medium",
Position = new ErrorPosition { Start = 0, End = 0 }
}).ToList();
return new GrammarCorrectionDto
{
HasErrors = true,
CorrectedText = string.Join(" ", corrections.Select(c => c.Correction)),
Corrections = corrections
};
}
private SentenceAnalysisData CreateFallbackAnalysis(string inputText, string aiResponse)
{
_logger.LogWarning("Using fallback analysis due to JSON parsing failure");
return new SentenceAnalysisData
{
OriginalText = inputText,
SentenceMeaning = aiResponse,
VocabularyAnalysis = new Dictionary<string, VocabularyAnalysisDto>(),
Metadata = new AnalysisMetadata
{
ProcessingDate = DateTime.UtcNow,
AnalysisModel = $"{_options.Model}-fallback",
AnalysisVersion = "2.0"
},
};
}
private void RecordSuccessfulRequest(long elapsedMs)
{
_stats.SuccessfulRequests++;
_stats.LastUsedAt = DateTime.UtcNow;
_stats.TotalCost += CostPerRequest;
UpdateAverageResponseTime((int)elapsedMs);
}
private void RecordFailedRequest(long elapsedMs)
{
_stats.FailedRequests++;
UpdateAverageResponseTime((int)elapsedMs);
}
private void UpdateAverageResponseTime(int responseTimeMs)
{
if (_stats.AverageResponseTimeMs == 0)
{
_stats.AverageResponseTimeMs = responseTimeMs;
}
else
{
_stats.AverageResponseTimeMs = (_stats.AverageResponseTimeMs + responseTimeMs) / 2;
}
}
#endregion
}
#region AI Response Models ()
internal class AiAnalysisResponse
{
public string? SentenceTranslation { get; set; }
public bool HasGrammarErrors { get; set; }
public List<AiGrammarCorrection>? GrammarCorrections { get; set; }
public Dictionary<string, AiVocabularyAnalysis>? VocabularyAnalysis { get; set; }
public List<AiIdiom>? Idioms { get; set; }
}
internal class AiGrammarCorrection
{
public string? Original { get; set; }
public string? Corrected { get; set; }
public string? Type { get; set; }
public string? Explanation { get; set; }
}
internal class AiVocabularyAnalysis
{
public string? Word { get; set; }
public string? Translation { get; set; }
public string? Definition { get; set; }
public string? PartOfSpeech { get; set; }
public string? Pronunciation { get; set; }
public string? DifficultyLevel { get; set; }
public string? Frequency { get; set; }
public List<string>? Synonyms { get; set; }
public string? Example { get; set; }
public string? ExampleTranslation { get; set; }
}
internal class AiIdiom
{
public string? Idiom { get; set; }
public string? Translation { get; set; }
public string? Definition { get; set; }
public string? Pronunciation { get; set; }
public string? DifficultyLevel { get; set; }
public string? Frequency { get; set; }
public List<string>? Synonyms { get; set; }
public string? Example { get; set; }
public string? ExampleTranslation { get; set; }
}
#endregion