Update 2024-02-02-textgen-improved-by-ctr.md

research/_posts/2024-02-02-textgen-improved-by-ctr.md

@@ -133,7 +133,7 @@ In order to improve the quality of the generated text, and to improve the RPM, w
 
 We first collect the results from ChatGPT to train our T5 model. We input product title and ask ChatGPT to write product description. The problems with the ChatGPT results are that the available rate of text is 89\% and 11\% of the product description is not suitable for display. So we clean the dataset based on ChatGPT API and train the T5 model with more than 99\% available rate of generated text. We use this generated results of our T5 model as the baseline, which is the Step-1 of Figure \cite{fig1}.
 
-
+We have tried putting the product attributes like product type and product tags into the prompts, combined with product title to query ChatGPT, but we do not observe some improvement of text quality and diversity by ChatGPT.
 
 ### 5. Relate Work