Update README.md

README.md

@@ -1,8 +1,8 @@
 ---
 title: KeyIntentNER T
 emoji: ♾️
-colorFrom: red
-colorTo: yellow
+colorFrom: yellow
+colorTo: red
 sdk: docker
 pinned: true
 license: apache-2.0
@@ -29,11 +29,29 @@ Utilizes GLiNER, an advanced Named Entity Recognition (NER) model, to classify s
 Matches keywords to topics from Google's well-known Content and Product taxonomies.
 
 ## Usage
-Enter a list of keywords (one per line, up to 100 MAX) and click the submit button. Keyword processing can take anywhere from 30 seconds up to ~2 minutes due to the extensive analysis performed behind the scenes. Once processing is complete, you can download any of the bar chart images and download a CSV export with insights for all keywords.
+- Enter a list of keywords (one per line, up to 100 MAX) and click the submit button. 
+- Keyword processing can take anywhere from 30 seconds up to ~2 minutes due to the extensive analysis performed behind the scenes. 
+- Once processing is complete, you can download any of the bar chart plots and download a CSV export with insights for all keywords.
+
+Example keywords: 
+```
+Standing desks vs. regular desks
+car repair service my area
+Buy groceries online with delivery
+average cost of car insurance
+top rated laptops 2024
+Book a flight to Hawaii
+how to fix a leaking faucet
+```
+
+#### Example Plot
+![KeyIntentNER-T_Example Plot](./images/keyintentner_t_example_plot.png)
+
+#### Example Plot
+![KeyIntentNER-T_Example Plot](./images/keyintentner_t_example_table.png)
 
 ## Benefits for SEO
-Improved Content Strategy
-Focus your SEO efforts on creating more relevant and helpful content that addresses the search intent for keywords.
+Improved content strategy by focusing your SEO efforts on creating more relevant/helpful content that addresses the search intent for keywords.
 
 ### Enhanced Keyword Targeting
 Match keywords to Google's well-known categories, ensuring your content is aligned with popular search themes.
@@ -41,10 +59,35 @@ Match keywords to Google's well-known categories, ensuring your content is align
 ### Better Understanding of User Intent
 Gain insights into what kind of information a person is looking for and how keywords can be interpreted by search engines.
 
+### Notes on Data:
+- GLiNER Model [gliner_small-v2.1](https://huggingface.co/urchade/gliner_small-v2.1) was trained on the [urchade/pile-mistral-v0.1](https://huggingface.co/datasets/urchade/pile-mistral-v0.1) dataset
+- Keyword Intent is determined by a custom function that looks for the presence of specific terms in keywords and classifies them into one of six predefined intent categories. The function includes popular "keyword modifiers" but a more customized list will provide better matches or using other ML tools to identify intent assignment. 
+- Keyword Intent Classification is performed using a custom rule-based algorithm. This algorithm scans keywords for specific terms and patterns, categorizing them into one of six predefined intent categories. While the current implementation includes common "keyword modifiers," the accuracy can be significantly improved by:
+  - Expanding and customizing the list of intent-indicating terms
+  - Implementing more sophisticated machine learning models for intent recognition
+- Topic Modeling uses both [Google Content Categories](https://cloud.google.com/natural-language/docs/categories) & [Google Product Taxonomy](https://www.google.com/basepages/producttype/taxonomy.en-US.txt)
+- GLiNER is able to use a large set of entity types, but for the purposes of this demo it is limited to the following [Google NLP Entity Types](https://cloud.google.com/natural-language/docs/reference/rest/v2/Entity#type):
+
+```
+PERSON : "person"
+LOCATION : "location"
+ORGANIZATION : "organization"
+EVENT : "event"
+WORK_OF_ART	: "work_of_art"
+CONSUMER_GOOD : "product" and "service"
+OTHER	: "misc"
+PHONE_NUMBER : "phone_number"
+ADDRESS	: "address"
+DATE	: "date"
+NUMBER : "number"
+PRICE : "price"
+```
+- Due to the limitations of hosting this in a free space, a smaller sentence transformers model is used which does not perform as well with some of the Topic Modeling categories. In testing, the [all-roberta-large-v1 model](https://huggingface.co/sentence-transformers/all-roberta-large-v1)  performed best for sample keywords tested. 
+  
 #### GLiNER Model Citation
 - GLiNER: Generalist Model for Named Entity Recognition using Bidirectional Transformer.
 - Authors: Urchade Zaratiana, Nadi Tomeh, Pierre Holat, Thierry Charnois.
 - Year: 2023.
 - Link: [arXiv:2311.08526](https://arxiv.org/abs/2311.08526)
 
-For questions or if you are interested in building custom SEO dash apps, contact me at: jrad.seo@gmail.com
+For questions or if you are interested in building custom SEO dash apps, contact me at: jrad.seo@gmail.com