Added seed phrases to KeyNMF #77
Conversation
x-tabdeveloping
changed the title
|
@KennethEnevoldsen can I has review? |
| @@ -20,42 +20,26 @@ | |||
| - Lemmatization and Stemming | |||
| - Visualization with [topicwizard](https://github.com/x-tabdeveloping/topicwizard) 🖌️ | |||
|
|
|||
| ## New in version 0.12.0: Seeded topic modeling | |||
There was a problem hiding this comment.
Seems like you should really keep a changelog (to many important tidbits in these that people likely miss out on)
You could potentially do it in a dropdown menu ("See previous versions (click to unfold)")
| @@ -8,20 +8,30 @@ while taking inspiration from classical matrix-decomposition approaches for extr | |||
| <figcaption>Schematic overview of KeyNMF</figcaption> | |||
| </figure> | |||
|
|
|||
|
|
|||
| Here's an example of how you can fit and interpret a KeyNMF model in the easiest way. | |||
There was a problem hiding this comment.
| Here's an example of how you can fit and interpret a KeyNMF model in the easiest way. | |
| Here's an example of how you can fit and interpret a KeyNMF model. |
| model.fit(corpus) | ||
|
|
||
| model.print_topics() | ||
| ``` | ||
|
|
||
| !!! question "Which Embedding model should I use" | ||
| - You should probably use KeyNMF with a `paraphrase-` type embedding model. These seem to perform best in most tasks. Some examples include: | ||
| - [paraphrase-MiniLM-L3-v2](https://huggingface.co/sentence-transformers/paraphrase-MiniLM-L3-v2) - Absolutely tiny :mouse: |
There was a problem hiding this comment.
I would focus on speed, not all will know that size and speed are related.
There was a problem hiding this comment.
Seems like this is a bit redundant if you can simply use the static-retrieval-mrl-en-v1?
| In KeyNMF, you can describe this aspect, from which you want to investigate your corpus, using a free-text seed-phrase, | ||
| which will then be used to only extract topics, which are relevant to your research question. |
There was a problem hiding this comment.
Has this idea been explored before? If so a reference would be great
| @@ -354,46 +424,49 @@ for batch in batched(zip(corpus, timestamps)): | |||
| model.partial_fit_dynamic(text_batch, timestamps=ts_batch, bins=bins) | |||
| ``` | |||
|
|
|||
| ### Hierarchical Topic Modeling | |||
| ## Asymmetric and Instruction-tuned Embedding Models | |||
There was a problem hiding this comment.
Some of these things are specifically in the KeyNMF docs, why not put them in a general section?
|
|
||
| ### Asymmetric and Instruction-tuned Embedding Models | ||
| ## Seeded Topic Modeling |
There was a problem hiding this comment.
Do you need to refer to the general documentation of seeded topic modeling here (seems like a lot of duplication).
Might it not be better to create a table like "supported types of topic modelling" and then in the "Seeded Topic Modelling" section add "Models which support seeded topic modelling".
| Some models are able to account for this by taking seed phrases or words. | ||
| This is currently only possible with KeyNMF in Turftopic, but will likely be extended in the future. | ||
|
|
||
| In [KeyNMF](../keynmf.md), you can describe the aspect, from which you want to investigate your corpus, using a free-text seed-phrase, |
There was a problem hiding this comment.
See the comment above as well.
Would probably write this more simply and then put this in a Tab section called KeyNMF (that way it is easy to see that the only one supported in KeyNMF, but also that there could be others in the future.
| @@ -120,6 +120,8 @@ def batch_extract_keywords( | |||
| self, | |||
| documents: list[str], | |||
| embeddings: Optional[np.ndarray] = None, | |||
| seed_embedding: Optional[np.ndarray] = None, | |||
| fitting: bool = True, | |||
There was a problem hiding this comment.
What fitting do?
You can now add a
seed_phraseto a KeyNMF model, which essentially indicates the aspect, from which the model has to examine documents.TODO: