这里会显示出您选择的修订版和当前版本之间的差别。
| 两侧同时换到之前的修订记录 前一修订版 后一修订版 | 前一修订版 | ||
|
zh:courses:textmining2026:ch03 [2026/09/14 16:02] pzczxs [上机作业] |
zh:courses:textmining2026:ch03 [2026/09/28 17:25] (当前版本) pzczxs [UTF-8文件工具类] |
||
|---|---|---|---|
| 行 1: | 行 1: | ||
| - | ====== 第三章 文本表示 ====== | + | ====== 第三章:文本表示 ====== |
| ===== 课件 ===== | ===== 课件 ===== | ||
| - | 下载:文本表示 | + | 下载:{{ :zh:courses:textmining2026:ch03.pptx |文本表示}} |
| 行 82: | 行 82: | ||
| } | } | ||
| </file> | </file> | ||
| - | |||
| ===== 基于Lucene的TF-IDF计算 ===== | ===== 基于Lucene的TF-IDF计算 ===== | ||
| ==== 第三方库依赖(POM.xml) ==== | ==== 第三方库依赖(POM.xml) ==== | ||
| 行 414: | 行 413: | ||
| </file> | </file> | ||
| + | ===== 向量模型 ===== | ||
| + | <file java OpenAICompatibleEmbeddingClient.java> | ||
| + | package cn.edu.bjut.textmining.chapter3; | ||
| + | |||
| + | import com.google.gson.JsonArray; | ||
| + | import com.google.gson.JsonElement; | ||
| + | import com.google.gson.JsonObject; | ||
| + | import com.google.gson.JsonParser; | ||
| + | |||
| + | import java.io.ByteArrayOutputStream; | ||
| + | import java.io.IOException; | ||
| + | import java.io.InputStream; | ||
| + | import java.io.OutputStream; | ||
| + | import java.io.UncheckedIOException; | ||
| + | import java.net.HttpURLConnection; | ||
| + | import java.net.URI; | ||
| + | import java.nio.charset.StandardCharsets; | ||
| + | import java.util.List; | ||
| + | |||
| + | /** 通过 OpenAI compatible API 获取文本向量。 */ | ||
| + | public final class OpenAICompatibleEmbeddingClient { | ||
| + | // 配置直接写在这里,无需设置环境变量。地址末尾不要添加 /embeddings。 | ||
| + | private static final String BASE_URL = | ||
| + | "https://llm-8olx338nm0p2vj5q.cn-beijing.maas.aliyuncs.com/compatible-mode/v1"; | ||
| + | |||
| + | // 这是占位用的假密钥;运行前在自己的电脑上替换为真实密钥。 | ||
| + | private static final String API_KEY = "sk-ws-H.PHL" ; | ||
| + | |||
| + | // qwen3.8-flash 是文本生成模型,无法用于本类的 /embeddings 向量请求。 | ||
| + | // 获取文本向量时,须填写服务支持的向量模型,如 qwen3.7-text-embedding, | ||
| + | // 并将 BASE_URL 和 API_KEY 一起换成有权调用该向量服务的配置。 | ||
| + | // 官方模型列表:https://help.aliyun.com/zh/model-studio/text-embedding-synchronous-api | ||
| + | private static final String MODEL = "qwen3.7-text-embedding"; | ||
| + | |||
| + | private final String baseUrl; | ||
| + | private final String apiKey; | ||
| + | private final String model; | ||
| + | |||
| + | public OpenAICompatibleEmbeddingClient(String baseUrl, String apiKey, String model) { | ||
| + | this.baseUrl = requireText(baseUrl, "BASE_URL").replaceAll("/+$", ""); | ||
| + | this.apiKey = requireText(apiKey, "API_KEY"); | ||
| + | this.model = requireText(model, "MODEL"); | ||
| + | URI uri = URI.create(this.baseUrl); | ||
| + | if (!("https".equals(uri.getScheme()) || "http".equals(uri.getScheme())) | ||
| + | || uri.getHost() == null || uri.getQuery() != null | ||
| + | || uri.getFragment() != null || uri.getUserInfo() != null) { | ||
| + | throw new IllegalArgumentException("BASE_URL 必须是 HTTP(S) API 根地址,不含查询参数"); | ||
| + | } | ||
| + | } | ||
| + | |||
| + | /** | ||
| + | * 供 EmbeddingApiDemo 调用。此方法直接使用上方常量,不读取环境变量。 | ||
| + | */ | ||
| + | public static OpenAICompatibleEmbeddingClient fromEnvironment() { | ||
| + | return new OpenAICompatibleEmbeddingClient(BASE_URL, API_KEY, MODEL); | ||
| + | } | ||
| + | |||
| + | public String getModel() { | ||
| + | return model; | ||
| + | } | ||
| + | |||
| + | /** 返回原始 JSON,调用者可将其保存后检查向量和请求信息。 */ | ||
| + | public String embed(List<String> texts) { | ||
| + | if (texts == null || texts.isEmpty()) { | ||
| + | throw new IllegalArgumentException("输入文本不能为空"); | ||
| + | } | ||
| + | JsonArray input = new JsonArray(); | ||
| + | for (String text : texts) { | ||
| + | input.add(requireText(text, "输入文本")); | ||
| + | } | ||
| + | JsonObject request = new JsonObject(); | ||
| + | request.addProperty("model", model); | ||
| + | request.add("input", input); | ||
| + | request.addProperty("encoding_format", "float"); | ||
| + | byte[] body = request.toString().getBytes(StandardCharsets.UTF_8); | ||
| + | try { | ||
| + | HttpURLConnection connection = (HttpURLConnection) | ||
| + | URI.create(baseUrl + "/embeddings").toURL().openConnection(); | ||
| + | // disconnect() 也在离开代码块时执行,包含请求失败的情况。 | ||
| + | try (AutoCloseable connectionResource = connection::disconnect) { | ||
| + | connection.setRequestMethod("POST"); | ||
| + | connection.setConnectTimeout(15000); | ||
| + | connection.setReadTimeout(60000); | ||
| + | connection.setInstanceFollowRedirects(false); | ||
| + | connection.setDoOutput(true); | ||
| + | connection.setRequestProperty("Authorization", "Bearer " + apiKey); | ||
| + | connection.setRequestProperty("Content-Type", "application/json; charset=UTF-8"); | ||
| + | connection.setRequestProperty("Accept", "application/json"); | ||
| + | connection.setFixedLengthStreamingMode(body.length); | ||
| + | try (OutputStream output = connection.getOutputStream()) { | ||
| + | output.write(body); | ||
| + | } | ||
| + | int status = connection.getResponseCode(); | ||
| + | boolean success = status >= 200 && status < 300; | ||
| + | String response = readResponse(success | ||
| + | ? connection.getInputStream() : connection.getErrorStream()); | ||
| + | if (!success) { | ||
| + | throw new IllegalStateException("向量 API 返回 HTTP " + status | ||
| + | + ":" + response.replace(apiKey, "[已隐藏密钥]")); | ||
| + | } | ||
| + | return response; | ||
| + | } | ||
| + | } catch (IOException e) { | ||
| + | throw new UncheckedIOException("请求向量 API 失败", e); | ||
| + | } catch (RuntimeException e) { | ||
| + | throw e; | ||
| + | } catch (Exception e) { | ||
| + | throw new IllegalStateException("关闭 HTTP 连接失败", e); | ||
| + | } | ||
| + | } | ||
| + | |||
| + | /** index 对应输入位置;响应数组的排列顺序不一定等于输入顺序。 */ | ||
| + | public static double[][] parseEmbeddings(String response, int inputCount) { | ||
| + | if (inputCount <= 0) { | ||
| + | throw new IllegalArgumentException("输入数量必须为正整数"); | ||
| + | } | ||
| + | try { | ||
| + | JsonArray data = JsonParser.parseString(response) | ||
| + | .getAsJsonObject().getAsJsonArray("data"); | ||
| + | if (data == null || data.size() != inputCount) { | ||
| + | throw new IllegalArgumentException("返回向量数量与输入数量不一致"); | ||
| + | } | ||
| + | double[][] vectors = new double[inputCount][]; | ||
| + | int dimension = -1; | ||
| + | for (JsonElement element : data) { | ||
| + | JsonObject item = element.getAsJsonObject(); | ||
| + | int index = item.get("index").getAsBigDecimal().intValueExact(); | ||
| + | if (index < 0 || index >= inputCount || vectors[index] != null) { | ||
| + | throw new IllegalArgumentException("响应 index 越界或重复:" + index); | ||
| + | } | ||
| + | JsonArray embedding = item.getAsJsonArray("embedding"); | ||
| + | if (embedding == null || embedding.size() == 0) { | ||
| + | throw new IllegalArgumentException("返回了空向量"); | ||
| + | } | ||
| + | if (dimension == -1) { | ||
| + | dimension = embedding.size(); | ||
| + | } | ||
| + | if (embedding.size() != dimension) { | ||
| + | throw new IllegalArgumentException("返回向量的维数不一致"); | ||
| + | } | ||
| + | vectors[index] = new double[dimension]; | ||
| + | double norm = 0.0; | ||
| + | for (int j = 0; j < dimension; j++) { | ||
| + | double value = embedding.get(j).getAsDouble(); | ||
| + | if (!Double.isFinite(value)) { | ||
| + | throw new IllegalArgumentException("向量中存在非有限数值"); | ||
| + | } | ||
| + | vectors[index][j] = value; | ||
| + | norm = Math.hypot(norm, value); | ||
| + | } | ||
| + | if (norm == 0.0 || !Double.isFinite(norm)) { | ||
| + | throw new IllegalArgumentException("向量范数必须是有限正数"); | ||
| + | } | ||
| + | } | ||
| + | for (double[] vector : vectors) { | ||
| + | if (vector == null) { | ||
| + | throw new IllegalArgumentException("响应缺少输入对应的向量"); | ||
| + | } | ||
| + | } | ||
| + | return vectors; | ||
| + | } catch (RuntimeException e) { | ||
| + | throw new IllegalArgumentException("无法解析有效的 embeddings 响应", e); | ||
| + | } | ||
| + | } | ||
| + | |||
| + | private static String readResponse(InputStream stream) { | ||
| + | if (stream == null) { | ||
| + | return ""; | ||
| + | } | ||
| + | try (InputStream input = stream; | ||
| + | ByteArrayOutputStream buffer = new ByteArrayOutputStream()) { | ||
| + | byte[] bytes = new byte[4096]; | ||
| + | for (int count; (count = input.read(bytes)) != -1; ) { | ||
| + | buffer.write(bytes, 0, count); | ||
| + | } | ||
| + | return new String(buffer.toByteArray(), StandardCharsets.UTF_8); | ||
| + | } catch (IOException e) { | ||
| + | throw new UncheckedIOException("读取向量 API 响应失败", e); | ||
| + | } | ||
| + | } | ||
| + | |||
| + | private static String requireText(String value, String name) { | ||
| + | if (value == null || value.trim().isEmpty()) { | ||
| + | throw new IllegalArgumentException("请设置非空的 " + name); | ||
| + | } | ||
| + | return value.trim(); | ||
| + | } | ||
| + | } | ||
| + | </file> | ||
| ===== 上机作业 ===== | ===== 上机作业 ===== | ||
| - | -给定一段文本(中文、英文均可),实现特征选择方法(熵、相对熵、$\xi^2$统计量和互信息),并完成特征选择。 | + | -给定一段文本(中文、英文均可),实现特征选择方法(熵、相对熵、$\chi^2$统计量和互信息),并完成特征选择。 |
| + | -利用第三方库,完成GloVe词嵌入的训练。 | ||
| ~~DISCUSSION~~ | ~~DISCUSSION~~ | ||