Air quality maps are crucial for urban environmental management, but the sparse deployment of Air Quality Monitoring Stations (AQMSs) leaves many areas unmonitored, requiring inference from limited observations. AQMS placement significantly impacts inference accuracy but remains underexplored. Consequently, we propose Greedy Batch Mode Query (GBMQ) for recommending optimal AQMS placement locations. GBMQ follows an active learning paradigm that recommends locations with high informativeness, characterized by high uncertainty (inference difficulty) and high impact (potential overall accuracy improvement). Uncertainty is estimated by Monte-Carlo Dropout, a Bayesian method suitable for both classification and regression tasks. While impact is assessed based on similarity in the learned feature space, offering a more reliable measure than raw feature similarity. Operating in batch mode, GBMQ effectively reduces computational cost and mitigates overfitting. And unlike existing methods that focus on a single pollutant, GBMQ simultaneously considers multiple pollutants when making recommendations, aligning with the multi-sensor nature of AQMSs. Experimental results demonstrate that GBMQ outperforms baseline methods on both NO \(_2\) and O \(_3\) maps, highlighting its effectiveness and generalizability.