Microservice autoscaling is crucial for maintaining the elasticity and cost efficiency of service architectures. While existing autoscaling approaches have made progress, they suffer from three interrelated limitations: First, they often rely on single metrics, failing to integrate multi-metric into coherent trend analyses that reveal holistic system behavior. Second, they typically do not expose the learned service dependencies or metric trends that underpin their predictions, making it difficult for users to understand the rationale behind scaling decisions. Furthermore, although some methods capture temporal dependencies, they fail to model the relative importance of different services in the spatial dimension. To bridge these gaps, we present VisScaler, a user-oriented visualization system designed to enhance microservice autoscaling by offering intuitive service topology visualizations and multi-metric trend analysis. Specifically, VisScaler extends the graph neural network (GNN)-based autoscaling model by supporting four heterogeneous input metrics (e.g., pod count, cpu utilization, response time, requests per second) and incorporating a node-level attention mechanism to model heterogeneous service importance, thereby improving prediction accuracy. Experimental results demonstrate that VisScaler outperforms state-of-the-art baselines, reducing the average MAE by 30.5% and the pod-count MAE by 16.0%. VisScaler is publicly accessible at \url{https://sqdcpc-blueeye.hf.space/}. The demonstration video is available at \url{https://youtu.be/EDrX3PWNgoE}.