Rapid estimation of the intensity of seismic ground motions is crucial for an effective rapid response when an earthquake occurs. To this end, maps of updated grond-motion fields (or shakemaps) are produced by using observations or measurements in near real-time to better constrain initial estimates. In this work, two types of observations are integrated to generate shakemaps right after an earthquake: the common type of data recorded by physical sensors (seismic stations) and the data extracted from social sensors (Twitter), or the combination of both. We investigate an approach to extract an approximation of the macroseismic intensity from social sensors 10 min after the earthquake; the approach relies on Twitter feeds to define the "felt area" where the earthquake was felt by the population, and the "unfelt locations" where the earthquake was not reported. Two recent earthquakes in France of moderate magnitude are studied and the results are compared to the official macroseismic intensity maps for validation. For the two studied cases, we note that Peak Ground Acceleration recordings far from the epicenter tend to underestimate the entire macroseismic field, and that the tweets from "felt areas" are complementary for a better estimation of the intensity shakemap. We highlight the importance and the limits of each type of observations when generating the seismic shakemaps.