A reasonable value assessment method plays an important role in both companies’ internal management and their external transactions. Traditional value assessment methods like the relative valuation method need historical financial indicators and some market multipliers but ignore non-financial information included in the text. Startup companies are usually listed before they start to make a profit, which means traditional value assessment methods are unable to access reliable financial information. This study firstly investigates companies’ non-financial information and proposes a text-based Enterprise Value Assessment System (EVAS) to extract features. Then, two neural network-based models are constructed to predict one essential factor K in order to improve the P/B relative valuation method. Our experiment in China's National Equities Exchange and Quotations (NEEQ) shows an impressive result on companies’ value assessment compared to the traditional value assessment model. The result concludes that introducing non-financial text information into value assessment based on our system improves the accuracy of value assessment for startup companies.