有关文本分析 API 的常见问题解答 (FAQ)Frequently Asked Questions (FAQ) about the Text Analytics API

查找与 Azure 认知服务中文本分析 API 有关的概念、代码和方案相关的常见问题解答。Find answers to commonly asked questions about concepts, code, and scenarios related to the Text Analytics API in Azure Cognitive Services.

文本分析能否识别嘲讽?Can Text Analytics identify sarcasm?

分析针对积极消极情绪,而不是进行情绪检测。Analysis is for positive-negative sentiment rather than mood detection.

情绪分析总是存在一定程度的不精确性,但是当内容没有隐藏的含义或潜台词时,该模型最有用。There is always some degree of imprecision in sentiment analysis, but the model is most useful when there is no hidden meaning or subtext to the content. 反讽、嘲讽、幽默和类似的微妙内容都依赖于文化背景和规范来传达意向。Irony, sarcasm, humor, and similarly nuanced content rely on cultural context and norms to convey intent. 这种类型的内容是最难分析的。This type of content is among the most challenging to analyze. 通常,分析器产生的给定分数与人的主观评估之间的最大差异在于具有微妙含意的内容。Typically, the greatest discrepancy between a given score produced by the analyzer and a subjective assessment by a human is for content with nuanced meaning.

我可以添加自己的训练数据或模型吗?Can I add my own training data or models?

不可以,模型是预先训练的。No, the models are pretrained. 对上传数据的唯一可用操作是评分、关键短语提取和语言检测。The only operations available on uploaded data are scoring, key phrase extraction, and language detection. 我们不托管自定义模型。We do not host custom models. 如果想要创建并托管自定义机器学习模型,请考虑 Microsoft R Server 中的机器学习功能If you want to create and host custom machine learning models, consider the machine learning capabilities in Microsoft R Server.

我可以请求其他语言吗?Can I request additional languages?

情绪分析和关键短语提取可用于部分语言Sentiment analysis and key phrase extraction are available for a select number of languages. 自然语言处理很复杂,需要进行大量测试才能发布新功能。Natural language processing is complex and requires substantial testing before new functionality can be released. 出于这个原因,我们避免预先宣布支持,这样就不会有人依赖需要更多时间才能成熟的功能。For this reason, we avoid pre-announcing support so that no one takes a dependency on functionality that needs more time to mature.

为帮助我们确定下一步要处理哪些语言的优先顺序,请在用户之声上为特定语言投票。To help us prioritize which languages to work on next, vote for specific languages on User Voice.

为什么关键短语提取会返回某些单词而不返回其他单词?Why does key phrase extraction return some words but not others?

关键短语提取消除了非必要词和独立形容词。Key phrase extraction eliminates non-essential words and standalone adjectives. 形容词 - 名词组合(例如“壮观的景色”或“有雾的天气”)将一起返回。Adjective-noun combinations, such as "spectacular views" or "foggy weather" are returned together.

通常,输出由名词和句子的宾语组成。Generally, output consists of nouns and objects of the sentence. 输出按重要性顺序列出,第一个短语是最重要的。Output is listed in order of importance, with the first phrase being the most important. 重要性按提及特定概念的次数或该元素与文本中其他元素的关系来衡量。Importance is measured by the number of times a particular concept is mentioned, or the relation of that element to other elements in the text.

为什么给定相同的输入,输出会不同?Why does output vary, given identical inputs?

如果更改较大,则会宣布对模型和算法进行改进;如果更新很小,则会悄悄地将其整合到服务中。Improvements to models and algorithms are announced if the change is major, or quietly slipstreamed into the service if the update is minor. 随着时间的推移,你可能会发现相同的文本输入会产生不同的情绪分数或关键短语输出。Over time, you might find that the same text input results in a different sentiment score or key phrase output. 这是在云中使用托管机器学习资源的正常且有意的结果。This is a normal and intentional consequence of using managed machine learning resources in the cloud.

后续步骤Next steps

问题是否与缺少功能相关?Is your question about a missing feature or functionality? 请考虑在我们的 UserVoice 网站上为其发起请求或投票。Consider requesting or voting for it on our UserVoice web site.

另请参阅See also