RASA NLU-我想在一个单词后面提取任何东西（单词，数字或特殊字符）作为一个实体

Question

Is there a way we can extract anything after a word as an entity; for eg:

I want to extract anything after about or go to or learn as an entity.

##intent:navigate
-I want to learn about linear regression
-I want to read about SVM
-I want to go to Python 2.6
-Take me to logistic regression: eval

##regex:topic
-^[A-Za-z0-9 :_ -][A-Za-z0-9 :_ -][A-Za-z0-9 :_ -]$

Answer 1

Naive way could be very simple - to use split string method eg

sentences = ["I want to learn about linear regression", "I want to read about SVM", "I want to go to Python 2.6",
 "Take me to logistic regression: eval"]

split_terms = ["about", "go to", "learn"]

for sentence in sentences:
    for split_term in split_terms:
        try:
            print(sentence.split(split_term)[1])
        except IndexError:
            pass # split_term was not found in a sentence

Results:

 linear regression
 about linear regression
 SVM
 Python 2.6

A little bit smarter way could be to find first the very last "splitting term" to solve issue with learn - learn about - about

for sentence in sentences:
    last_split_term_index = 0
    last_split_term = ""
    for split_term in split_terms:
        last_split_term_index_candidate = sentence.find(split_term)
        if last_split_term_index_candidate > last_split_term_index:
            last_split_term_index = last_split_term_index_candidate
            last_split_term = split_term
    try:
        print(sentence.split(last_split_term)[1])

    except:
        continue

Results:

 linear regression
 SVM
 Python 2.6

Answer 2

Yes you can, you will have to define the entities in your training data and it would be extracted by the model. For instance, in your example training data should be like.

##intent:navigate
- I want to learn about [linear regression](topic)
- I want to talk about [RasaNLU](topic) for the rest of the day.
- I want to go to [Berlin](topic) for a specific work.
- I want to read about [SVM](topic)
- I want to go to [Python 2.6](topic)
- Take me to logistic regression: eval

After model training I tried an example

Enter a message: I want to talk about SVM     
{
  "intent": {
    "name": "navigate",
    "confidence": 0.9576369524002075
  },
  "entities": [
    {
      "start": 21,
      "end": 24,
      "value": "SVM",
      "entity": "topic",
      "confidence": 0.8241770362411013,
      "extractor": "CRFEntityExtractor"
    }
  ]
}

But for this to be effective you will have to define more examples with all possible patterns. Like the example 'I want to talk about RasaNLU for the rest of the day.' suggests the model that the entity to be extracted doesn't have to be the last word of the sentence (which is the case in rest of the examples).