Conventionally, copy correction in various schools and colleges is done either by teachers or teacher assistants which is usually a tedious process and sometimes biased.To simplify this tedious task and make the system non-partial, we developed a software in which using OCR , scanned images of students answer sheets will be analysed and matched with model answers using NLP and generate marks accordingly.
The major challenge was Optically recgonizing text and in what respect measure similarity between student's answer and model answer. We came up with the below solution.The images of model answers and students answers are uploaded to google drive and are then converted to doc format and only the OCR recognised text is downloaded.All this is done via Google Drive API. Now our data is ready for evaluation.We are generating keywords and wordmappings through sentences which we will be read from previously generated text files. We will generate percentage scores by applying principles of semantic similarity on the keywords and mappings generated above. The results are then displayed in a tabular form.
Technologies used
Discussion