# Rasa\_NLU ner\_crf classification issue

**URL:** <https://forum.rasa.com/t/rasa-nlu-ner-crf-classification-issue/10536>\
**Category:** Rasa Open Source\
**Created:** [May 28, 2019, 2:19pm UTC](https://forum.rasa.com/t/rasa-nlu-ner-crf-classification-issue/10536 "2019-05-28T14:19:29Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![Kenz](https://avatars.discourse-cdn.com/v4/letter/k/ee7513/32.png) [@Kenz](https://forum.rasa.com/u/Kenz)\
**Post date:** [May 28, 2019, 2:19pm UTC](https://forum.rasa.com/t/rasa-nlu-ner-crf-classification-issue/10536/1 "2019-05-28T14:19:29Z")

</div>

Hi, I’m currently building a chatbot using Rasa-NLU and using ner\_crf as entity classifier in the pipeline.

I’m having around half a million training sentences with only 12 different entities. The extraction is going well but the recognition is not that accurate…

I’m trying to find why…

The pipeline is as folow: language: “fr”

pipeline:

- name: “components.preprocess.PrepareString”
- name: “nlp\_spacy”
- name: “tokenizer\_spacy”
- name: “ner\_crf” features: [[“low”], [“bias”, “suffix3”], [“upper”, “pos”, “pos2”]]
- name: “ner\_synonyms”
- name: “intent\_featurizer\_count\_vectors”
- name: “intent\_classifier\_tensorflow\_embedding”

I believe that it may be due to my none understanding of the features on ner\_crf… Could someone explain to me what are the different features for ?

For example:

- low
- title
- suffix5
- suffix3
- suffix2
- suffix1
- pos
- pos2
- prefix5
- prefix2
- bias
- upper
- digit

---

<div class="post-metadata">

**Author:** ![Ghostvv](https://dub1.discourse-cdn.com/flex013/user_avatar/forum.rasa.com/ghostvv/32/211_2.png) [@Ghostvv](https://forum.rasa.com/u/Ghostvv)\
**Post date:** [June 12, 2019, 5:08pm UTC](https://forum.rasa.com/t/rasa-nlu-ner-crf-classification-issue/10536/2 "2019-06-12T17:08:37Z")

</div>

what do you mean by?

> [@Kenz](#):
>
> The extraction is going well but the recognition is not that accurate…