The following contains a high-level view of what will be the next main enhancement steps. This document will be kept up-to-date and improved frequently. This work will be mainly conducted by @mk2510 and @henrifroese as part of their SummerOfCode project.

1. Version 1.10

   • Every representation function to receive as input a TokenSeries TokenSeries as input to every representation function #44
   • Decouple TF-IDF L2-normalization and TF-IDF tfidf(s): remove normalization, improve docstring #76
   • Rename term_frequency to count() + add functionterm_frequency count(s) and term_frequency(s) #61
   • Introduce HeroSeries
   • Add ~ hero.norm(RepresetationSeries, "l1"/"l2")
   • Can we avoid the use of VectorSeries/TokenSeries?
   • All representation functions to deal with HeroSeries + (DocumentTermDF) Support "Pandas Series Representation" #43
   • Update README + getting-started.md
   • Push a new version to PyPi

2. Performance: speed-up the library

   • Most of Texthero data structure are list of list ([["a", "document"], ["another", "document"]]), can we leverage parallelization? We can learn from spaCy. Mandatory read: 100-times-faster-nlp; look at this for parallelization
   • Make spaCy function faster + Dask vs Spacy Make spaCy-nlp functions faster #65
   • Depending on the previous task, evaluate if we want to have as default tokenizer spaCy: tokenize with Spacy #131

3. Software development:

   • Integrate checking for correct Series types (Kind of Pandas Series #60, Check if Series consists of strings only, instead of casting to unicode #55, ...)
   • Check hero functions work with np.nan All function to deal with np.nan #86

4. Support Embeddings through Flair

   • Add hero.embed(s, flairEmbedding)

5. Add Topic Modeling

   • Add topic modeling support under representation Implement/support/explain topic modelling #42
     This include also "topic modeling visualization" to get insights out of it
   • Add a blog article on how topic modeling with Texthero works

6. Extra

   1. test coverage
   2. expand multilingual: more languages; recognize languages and select correct one
   3. (low priority) Text summarization (Preprocessing: explain how to create a custom pipeline #38) and characteristic terms (Characteristic Terms & Keywords #2)