// btnPredict_Click 事件处理方法
    protected void btnPredict_Click(object sender, EventArgs e)
    {
        try
        {
            // 从文本框中获取输入文本
            string inputText = txtInput.Text.Trim();

            // 加载 Tokenizer 和 LabelEncoder 数据
            var tokenizer = LoadTokenizer(Server.MapPath("文件资料/JSON文件/word_index.json"));
            var labelEncoder = LoadLabelEncoder(Server.MapPath("文件资料/JSON文件/label_encoder.json"));

            // 1. 文本预处理
            var processedText = PreprocessText(inputText, tokenizer);

            var options = new SessionOptions();


            // 设置日志级别为信息级别
            options.LogSeverityLevel = OrtLoggingLevel.ORT_LOGGING_LEVEL_INFO;

            // 2. 加载 ONNX 模型
            var session = new InferenceSession(Server.MapPath("onnx_file/my_lstm_model.onnx"),options);
            
            // 3. 创建输入张量
            var inputName = session.InputMetadata.Keys.First(); // 获取输入名称
            var inputTensor = new DenseTensor<float>(processedText, new[] { 1, processedText.Length });

            // 4. 进行推理
            var namedOnnxValue = NamedOnnxValue.CreateFromTensor(inputName, inputTensor);
            var results = session.Run(new[] { namedOnnxValue });

            // 5. 获取预测结果
            var prediction = results.First().AsTensor<float>().ToArray();
            var predictedClassIndex = Array.IndexOf(prediction, prediction.Max());

            // 6. 使用 LabelEncoder 进行反编码
            string predictedClass = labelEncoder[predictedClassIndex];

            // 7. 显示预测结果
            lblResult.Text = "预测结果: " + predictedClass;
            
        }
        catch (Exception ex)
        {
            // 处理可能出现的错误
            lblResult.Text = "错误: " + ex.Message;
        }
    }

    // 文本预处理
    static float[] PreprocessText(string text, Dictionary<string, int> tokenizer)
    {
        var words = TokenizeText(text);  // 分词
        var sequence = words.Select(word => tokenizer.ContainsKey(word) ? tokenizer[word] : 0).ToArray();
        var paddedSequence = PadSequence(sequence, maxLength: 50);
        return paddedSequence.Select(x => (float)x).ToArray();
    }

    // 使用 jieba.net 进行分词
    static List<string> TokenizeText(string text)
    {
        var segmenter = new JiebaSegmenter();
        var words = segmenter.Cut(text).ToList();
        return words;
    }

    // 填充序列
    static int[] PadSequence(int[] sequence, int maxLength)
    {
        var padded = new int[maxLength];
        for (int i = 0; i < Math.Min(sequence.Length, maxLength); i++)
        {
            padded[i] = sequence[i];
        }
        for (int i = sequence.Length; i < maxLength; i++)
        {
            padded[i] = 0;
        }
        return padded;
    }

    // 加载 Tokenizer
    static Dictionary<string, int> LoadTokenizer(string path)
    {
        var tokenizerJson = File.ReadAllText(path);
        return JsonConvert.DeserializeObject<Dictionary<string, int>>(tokenizerJson);
    }

    // 加载 LabelEncoder 类别
    static string[] LoadLabelEncoder(string path)
    {
        var labelEncoderJson = File.ReadAllText(path);
        return JsonConvert.DeserializeObject<string[]>(labelEncoderJson);
    }
}