How to understand the method of least squares? ( Despised not know what the basic principles of statistics? This article was written in nice easy to understand )
https://blog.csdn.net/ccnt_2012/article/details/81127117
Detailed Python installation directory structure ( with PYTHON still need to be familiar environment under WINDOWS )
https://blog.csdn.net/qq_43401808/article/details/88869397
Python + wordcloud + jieba + docx generation Chinese word cloud and word frequency statistics ( that was fun )
https://blog.csdn.net/fengjianc/article/details/78929121
Dialogue on understanding people with machines
Speech Act Theory speech act theory of
Austin believes that any word, people at the same time to complete the three acts: locutionary act, illocutionary act, perlocutionary act (Guyun Ying, 1986)
Austin illocutionary acts divided into five categories: the determining language (verdictives), ruled that language (exercitives), commitment language (commissives), elaborated language (expositives) and behavior language (behabitives). Later, Searle criticized this classification, the illocutionary act into "new five": that is, the assertion (assertives), instructions (directives), commitment language (commissives), expression language (expressives) and announced (declarations).
Understanding human questions
after Intel survey found that for Chinese problem, it is nothing more than can be divided into two categories: questions and rhetorical questions. For the rhetorical question of course, nothing to say, let's look at the focus questions. It can be divided into non-questions, negative questions, especially questions, alternative questions, especially questions which can be divided into people, cause, location, time, advice, quantity, and the way the rest of the entity.
For questions, humans also need to make a judgment on the sentence, especially to get the problem, the need to determine in the end is what to ask? Each question will then make an initial position, when the answer to narrow your search, finally get answers from the knowledge and scene.
The main difficulty in the corpus of NLU ready, then one by one recorded on his own experience learned.
Each keyword is required intent, the intent of the sentence should have a keyword.
Each keyword to expand about 20 statements.
All statements should be enough divergence between discrete (that is, except the keyword try not to repeat the words).
In addition to keywords, all words word repetition rate is lower in each of intent, is lower, it is best not to repeat.
Throughout this document, in addition to keywords, all words word repetition rate is lower, much lower, it is best not to repeat.
The above phenomenon is caused by two, you and me ah yes you have to get rid of such words (semantics can be a little incoherently, acceptable).
The same sentence, different parameters intended to be combined and resolved by post-calibration parameter.
Intention recognition accuracy with both the relevant
Frequency keyword appears in the current intention of the
frequency of the keyword that appears throughout the document
Original | industrial scene, quizzes How chatting robot?
https://www.sohu.com/a/270483785_473476
the NLU (Natural Language Understanding) is the use of NLP (Natural Language Processing) technology intended to identify problems and user entity extraction. Identifying user intent is to figure out what in the end to ask, cause of the malfunction or the number of inquiries the case of failure; entity extraction is a specific slot value of this intention. Questions such as "What is the last month the number of generator failure," the intent is to "Query number of failures," the value of slot fault name is "generator failure", the value of time slots is "last month." Can be described as intended to identify classification issues, the use of machine learning methods to solve, such as SVM, fastText; entity extraction using NLP in the NER (NER) related technology solutions.
Rasa Guide 01 ( think of ways to do a robot out? )
https://terrifyzhao.github.io/2018/09/17/Rasa%E4%BD%BF%E7%94%A8%E6%8C%87%E5%8D%9701.html
https://terrifyzhao.github.io/2019/02/26/Rasa%E4%BD%BF%E7%94%A8%E6%8C%87%E5%8D%9702.html
Natural language processing sequence of labeling problem ( with a machine understandable way to deal with the problem )
https://www.cnblogs.com/jiangxinyang/p/9368482.html
Detailed article apply deep learning in NER ner in ( in the end is the first word or NER do first tangled for a long time, and read not tangle )
NER system concept carding
NER major categories, generally include three categories (based entity, digital time-based and the like) and 7 subclasses (place names, organization name, organization name, time, date, money and percentages). But with the continued expansion of NLP tasks, there will be specific category in specific areas, such as the field of medicine, the name of the drug, disease, and other categories.
https://blog.csdn.net/f12105212/article/details/78500327
Master switch from the material to the road of algorithm engineers ( this guy looks very Niua, to learn from him, very polite people, but also added a WX )
https://www.cnblogs.com/jiangxinyang/p/10263414.html
