斯坦福 NLP 情绪模糊结果

Question

我正在使用 Stanford NLP v3.6 (JAVA) 来计算 English 句子的 sentiment。

Stanford NLP 从 0 到 4 计算句子的极性。

• 0 非常负面
• 1 否定
• 2 中立
• 3 阳性
• 4 非常积极

我运行了一些非常简单的测试用例，但得到了非常奇怪的结果。

例子：

1. Text = Jhon 是好人，Sentiment = 3（即积极）
2. Text = David 是个好人，Sentiment = 2（即中性）

在上面的例子中，除了名字David，Jhon之外，句子都是一样的，但是情感值不同。 这不是模棱两可吗？

我使用这个 Java 代码来计算情绪：

 public static float calSentiment(String text) {

            // pipeline must get initialized before proceeding further
            Properties props = new Properties();
            props.setProperty("annotators", "tokenize, ssplit,   parse, sentiment");
            StanfordCoreNLP pipeline = new StanfordCoreNLP(props);

            int mainSentiment = 0;
            if (text != null && text.length() > 0) {
                int longest = 0;
                Annotation annotation = pipeline.process(text);

                for (CoreMap sentence : annotation.get(CoreAnnotations.SentencesAnnotation.class)) {
                    Tree tree = sentence.get(SentimentCoreAnnotations.SentimentAnnotatedTree.class);
                    int sentiment = RNNCoreAnnotations.getPredictedClass(tree);
                    String partText = sentence.toString();

                    if (partText.length() > longest) {
                        mainSentiment = sentiment;
                        longest = partText.length();
                    }
                }
            }
            if (mainSentiment > 4 || mainSentiment < 0) {
                return -9999;
            }
            return mainSentiment;

        }

我在 java 代码中是否遗漏了什么？

当句子是积极的时候，我也得到消极情绪（即小于 2），反之亦然。

谢谢。

以下是我用简单的英文句子得到的结果：

Sentence: Tendulkar is a great batsman
Sentiment: 3
Sentence: David is a great batsman
Sentiment: 3
Sentence: Tendulkar is not a great batsman
Sentiment: 1
Sentence: David is not a great batsman
Sentiment: 2
Sentence: Shyam is not a great batsman
Sentiment: 1
Sentence: Dhoni loves playing football
Sentiment: 3
Sentence: John, Julia loves playing football
Sentiment: 3
Sentence: Drake loves playing football
Sentiment: 3
Sentence: David loves playing football
Sentiment: 2
Sentence: Virat is a good boy
Sentiment: 2
Sentence: David is a good boy
Sentiment: 2
Sentence: Virat is not a good boy
Sentiment: 1
Sentence: David is not a good boy
Sentiment: 2
Sentence: I love every moment of life
Sentiment: 3
Sentence: I hate every moment of life
Sentiment: 2
Sentence: I like dancing and listening to music
Sentiment: 3
Sentence: Messi does not like to play cricket
Sentiment: 1
Sentence: This was the worst movie I have ever seen
Sentiment: 0
Sentence: I really appreciated the movie
Sentiment: 1
Sentence: I really appreciate the movie
Sentiment: 3
Sentence: Varun talks in a condescending way
Sentiment: 2
Sentence: Ram is angry he did not win the tournament
Sentiment: 1
Sentence: Today's dinner was awful
Sentiment: 1
Sentence: Johny is always complaining
Sentiment: 3
Sentence: Modi's demonetisation has been very controversial and confusing
Sentiment: 1
Sentence: People are left devastated by floods and droughts
Sentiment: 2
Sentence: Chahal did a fantastic job by getting the 6 wickets
Sentiment: 3
Sentence: England played terribly bad
Sentiment: 1
Sentence: Rahul Gandhi is a funny man
Sentiment: 3
Sentence: Always be grateful to those who are generous towards you
Sentiment: 3
Sentence: A friend in need is a friend indeed
Sentiment: 3
Sentence: Mary is a jubilant girl
Sentiment: 2
Sentence: There is so much of love and hatred in this world
Sentiment: 3
Sentence: Always be positive
Sentiment: 3
Sentence: Always be negative
Sentiment: 1
Sentence: Never be negative
Sentiment: 1
Sentence: Stop complaining and start doing something
Sentiment: 2
Sentence: He is a awesome thief
Sentiment: 3
Sentence: Ram did unbelievably well in this year's exams
Sentiment: 2
Sentence: This product is well designed and easy to use
Sentiment: 3

Answer 1

情绪决策由经过训练的神经网络做出。不幸的是，根据您在同一个句子中提供的不同名称，它的行为很奇怪，但这是意料之中的。正如 GitHub 上所讨论的，一个因素可能是各种名称在训练数据中不经常出现。