Phase 05: NLP: Foundations to Advanced

感觉分析

关于经典文本分类的大部分知识都在这里.

Type: Build

Languages: Python

Prerequisites: Phase 5 · 02 (BoW + TF-IDF), Phase 2 · 14 (Naive Bayes)

Time: ~75 minutes

问题

"食物不好". 积极还是消极?

感觉听起来很简单.一个评论员说他们喜欢或不喜欢某种东西.标签句子.它成为了神圣的NLP任务的原因是每个容易看起来的案例都隐藏着一个难以理解的.否定翻转意义.刺反转意义. "不坏的"是积极的,尽管有两个负面编码的词.爱莫吉带有比周围文本更多的信号.域名词汇问题 (tight在音乐评论中tight在时尚审查中).

感觉是经典NLP的工作实验室.如果你明白为什么每一个天真的基线都有特定的失败模式,你就会明白为什么每一个更丰富的模型都被发明了.这个课程从零开始构建了一个天真的贝叶斯基线,添加了物流回归,并命名了使生产感觉成为合规程度问题的陷.

概念

经典情感是一个两步的食谱.

Represent.转换文本为特征向量.
Classify.根据标记的例子,适应线性模型 (Naive Bayes,物流回归,SVM).

简单的贝耶斯是最愚蠢的模型,假设每个特征都独立,P(word | positive)其他P(word | negative)根据"无常"独立假设,这是一个可笑的错误,但结果却令人震惊.原因是:由于文本的特征稀少,并且数据中等,分类器关心每个词的倾向是哪个方面,而不是多少.

逻辑回归修复了独立假设. 它学习每个特征的权重,包括负权重. not good简单的贝耶斯不能为它从未标记过的比格拉姆做.

建立它

步骤1:一个真正的微型数据集

pythonPOSITIVE = [
    "absolutely loved this movie",
    "beautiful cinematography and a great story",
    "one of the best films of the year",
    "brilliant acting from the lead",
    "heartwarming and funny",
]

NEGATIVE = [
    "boring and far too long",
    "not worth your time",
    "the plot made no sense",
    "terrible acting, awful script",
    "i want my two hours back",
]

实际工作使用了数万个例子 (IMDb,SST-2,Yelp极度).数学是相同的.

步骤2:从零开始,多个个字母的天真贝耶斯

pythonimport math
from collections import Counter


def train_nb(docs_by_class, vocab, alpha=1.0):
    class_priors = {}
    class_word_probs = {}
    total_docs = sum(len(d) for d in docs_by_class.values())

    for cls, docs in docs_by_class.items():
        class_priors[cls] = len(docs) / total_docs
        counts = Counter()
        for doc in docs:
            for token in doc:
                counts[token] += 1
        total = sum(counts.values()) + alpha * len(vocab)
        class_word_probs[cls] = {
            w: (counts[w] + alpha) / total for w in vocab
        }
    return class_priors, class_word_probs


def predict_nb(doc, class_priors, class_word_probs):
    scores = {}
    for cls in class_priors:
        s = math.log(class_priors[cls])
        for token in doc:
            if token in class_word_probs[cls]:
                s += math.log(class_word_probs[cls][token])
        scores[cls] = s
    return max(scores, key=scores.get)

没有它,一个词在类中看不到的概率为零,并且日志爆炸. alpha=0.01实际上,这种情况是常见的.alpha=1.0现在,我们在教学中,

步骤3:从零开始的物流回归

pythonimport numpy as np


def sigmoid(x):
    return 1.0 / (1.0 + np.exp(-np.clip(x, -20, 20)))


def train_lr(X, y, epochs=500, lr=0.05, l2=0.01):
    n_features = X.shape[1]
    w = np.zeros(n_features)
    b = 0.0
    for _ in range(epochs):
        logits = X @ w + b
        preds = sigmoid(logits)
        err = preds - y
        grad_w = X.T @ err / len(y) + l2 * w
        grad_b = err.mean()
        w -= lr * grad_w
        b -= lr * grad_b
    return w, b


def predict_lr(X, w, b):
    return (sigmoid(X @ w + b) >= 0.5).astype(int)

文本特征很少,没有L2模型记住训练示例.0.01听,听听.

操作否定 (故障模式)

考虑"不好"和"不坏".{not, good}其他{not, bad}们的们都会看到一个大类别.not_good其他not_bad对于这些问题,我们需要了解更多的信息.

没有大子的时候可以做得更好.negation scoping后面的代码是否定字符.NOT_接下来的分字.

pythonNEGATION_WORDS = {"not", "no", "never", "nor", "none", "nothing", "neither"}
NEGATION_TERMINATORS = {".", "!", "?", ",", ";"}


def apply_negation(tokens):
    out = []
    negate = False
    for token in tokens:
        if token in NEGATION_TERMINATORS:
            negate = False
            out.append(token)
            continue
        if token in NEGATION_WORDS:
            negate = True
            out.append(token)
            continue
        out.append(f"NOT_{token}" if negate else token)
    return out

python>>> apply_negation(["not", "good", "at", "all", ".", "but", "funny"])
['not', 'NOT_good', 'NOT_at', 'NOT_all', '.', 'but', 'funny']

现在good其他NOT_good它们的分类器可以对比重重. 三行预处理,可测量的准确性跳跃于情感基准.

步骤5:重要的评估指标

只有在类别不平衡的情况下,准确性就会误导.实际的情感体通常是70-80%正确的或70-80%负面的;一个常数多数的分类器获得80%的准确性,并且是无价值的.报告以下每一个:

Per-class precision and recall.给每班一个对,对它们进行宏观平均,以得到一个尊重班级平衡的单个数字.
Macro-F1 (primary metric for imbalanced data).平均每类F1分数,均重.当类不平衡时,使用这个比较.
Weighted-F1 (alternative).报告与宏F1同时,当不平衡本身具有商业意义时.
Confusion matrix.总是检查之前信任任何规模度量; 它揭示模型混的类对.
Per-class error samples.每班就有五个错误预测,阅读它们. 没有什么可以取代读到实际错误.

对于严重失衡的数据 (> 95-5 比例),报告AUROC其他AUPRC非洲人民共和国人民共和国 (AUPRC) 对少数民族群体更敏感,这就是你通常关心的 (垃圾邮件,欺诈,罕见情绪).

Common bug to avoid.报告微F1而不是宏F1在不平衡数据上给出一个看起来很高的数字,因为它由多数类占主导地位.

pythondef evaluate(y_true, y_pred):
    tp = sum(1 for t, p in zip(y_true, y_pred) if t == 1 and p == 1)
    fp = sum(1 for t, p in zip(y_true, y_pred) if t == 0 and p == 1)
    fn = sum(1 for t, p in zip(y_true, y_pred) if t == 1 and p == 0)
    tn = sum(1 for t, p in zip(y_true, y_pred) if t == 0 and p == 0)
    precision = tp / (tp + fp) if tp + fp else 0
    recall = tp / (tp + fn) if tp + fn else 0
    f1 = 2 * precision * recall / (precision + recall) if precision + recall else 0
    return {"tp": tp, "fp": fp, "tn": tn, "fn": fn, "precision": precision, "recall": recall, "f1": f1}

用它

子学习用六行,正确.

pythonfrom sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.linear_model import LogisticRegression
from sklearn.pipeline import Pipeline

pipe = Pipeline([
    ("tfidf", TfidfVectorizer(ngram_range=(1, 2), min_df=2, sublinear_tf=True, stop_words=None)),
    ("clf", LogisticRegression(C=1.0, max_iter=1000)),
])
pipe.fit(X_train, y_train)
print(pipe.score(X_test, y_test))

需要注意的三个东西.stop_words=None没有任何证据.ngram_range=(1, 2)增加了大图.not_good成为一个特征.sublinear_tf=True它们是SST-2的75%准确基线和85%准确基线之间的区别.

什么时候要找变压器

刺的检测,古典模型失败了.
长期的评论,情绪在文件中转移.
基于面积的感觉. "相机很棒,但电池很糟糕".你需要把感觉归因于面积.
无英语,资源低.多语言BERT免费提供零截图的基础线.

如果您需要上述任何一个,请跳转到第7阶段 (变压器深入潜水).否则,TF-IDF+大图+否定处理的无知贝斯或物流回归将是2026年生产基线.

复制性陷 (再次)

重新训练情感模型是常规的.重新评估它们不是.报纸中报告的准确数量使用特定的分区,特定的预处理,特定的代币化.如果你比较你的新模型与一个基线,而不使用相同的管道,你会得到误导性地分数.总是重建你的基线,而不是纸质的数字.

运送它

保存如outputs/prompt-sentiment-baseline.md其他:

markdown---
name: sentiment-baseline
description: Design a sentiment analysis baseline for a new dataset.
phase: 5
lesson: 05
---

Given a dataset description (domain, language, size, label granularity, latency budget), you output:

1. Feature extraction recipe. Specify tokenizer, n-gram range, stopword policy (usually keep), negation handling (scoped prefix or bigrams).
2. Classifier. Naive Bayes for baseline, logistic regression for production, transformer only if the domain needs sarcasm / aspects / cross-lingual.
3. Evaluation plan. Report precision, recall, F1, confusion matrix, and per-class error samples (not just scalars).
4. One failure mode to monitor post-deployment. Domain drift and sarcasm are the top two.

Refuse to recommend dropping stopwords for sentiment tasks. Refuse to report accuracy as the sole metric when classes are imbalanced (e.g., 90% positive). Flag subword-rich languages as needing FastText or transformer embeddings over word-level TF-IDF.

运动

Easy.加入apply_negation作为一个预处理步骤,在 scikit-学习管道中测量F1三角形在一个小的情感数据集.
Medium.实施按类权重的物流回归 (通过 class_weight="balanced"测量对合成90-10类失衡的影响.
Hard.通过训练第二个分类器对情感模型的残余进行刺探测器. 记录你的实验设置. 当你的准确性低于机会时,警告读者 (二级刺的机会水平是50%左右,大多数第一次尝试都会降落在那里).

关键词

Term	What people say	What it actually means
Polarity	Positive or negative	Binary label; sometimes extended to neutral or fine-grained (5-star).
Aspect-based sentiment	Per-aspect polarity	Attribute sentiment to specific entities or attributes mentioned in text.
Negation scoping	Reversing nearby tokens	Prefix tokens after "not" with `NOT_` until punctuation.
Laplace smoothing	Adding 1 to counts	Prevents zero-probability features in Naive Bayes.
L2 regularization	Shrinking weights	Adds `lambda * sum(w^2)` to loss. Essential for sparse text features.

进一步阅读

Pang and Lee (2008). Opinion Mining and Sentiment Analysis长时间,但第一四节涵盖了古典的全部内容.
Wang and Manning (2012). Baselines and Bigrams: Simple, Good Sentiment and Topic Classification报纸显示了大图 +天真的贝耶斯,
scikit-learn text feature extraction docs参考CountVectorizer现在TfidfVectorizer你会调节每一个按.

This free lesson is part of the AI Engineering from Scratch curriculum. Read the full explanation, run the lesson code, and verify the result in the interactive reader or from the repository source.

Browse the complete course catalog or open this lesson on GitHub.