The proposal made in this paper is a strategic framework for advertisement recommendation as short videos. This approach employs a knowledge centric, semantically-inclined framework with two unique hybrid classification approaches, namely, Gated Recurrent Unit (GRU) and the decision tree classifier to classify the data generated in the dataset perspective as well as the querying perspective, respectively, the common classified and matched classes are reprioritized and the query words preprocessed and subjected to Latent Dirichlet Allocation (LDA) and distinct knowledge base encompassment through CYC and WikiData is sunk into the proposed framework. Semantics oriented reasoning and learning is achieved through semantics similarity computation using CoSimRank and Petriates index at different stages in the pipeline along with Mautista index for feature selection at the query end to classify the metadata in this invasive weed optimization has been encompassed to yield the best in class facets from which the corresponding advertisement videos are recommended with an overall precision of 95.03%, F-measure of 96.45 and false discovery rate (FDR) of 0.05 which makes the suggested framework the best in class for multimedia advertisement recommendations.