Abstract: With the acceleration of urbanization, the problem of urban traffic congestion has become increasingly serious. The traditional congestion analysis methods are difficult to meet the needs of refined management due to low sampling rate and high cost. Based on the real-time traffic data of Baidu map, this study proposes a "semantic+spatial" two-dimensional analysis framework based on online map: (1) Using natural language processing (NLP) technology to analyze the semantic description of road congestion and identify intersection congestion nodes; (2) Eliminate false congestion points caused by congestion spread based on road network topology analysis; (3) Build a multidimensional evaluation system covering congestion frequency, average congestion degree and congestion evaluation value; (4) According to the distribution of congestion directions, the intersections are divided into three types: single direction congestion, two-way congestion on trunk roads and multi-directional congestion, and the traffic congestion control strategies are given. Taking Haizhu District of Guangzhou as an example, it is found that the congestion in this area is characterized by obvious peak period, and the main congestion is concentrated in the North-south transit traffic in the western area; Among all the congestion points, single direction congestion accounts for the largest proportion, which can be alleviated by micro reconstruction measures such as signal optimization and lane adjustment; Most of the nodes with high congestion evaluation value are multi-directional congestion, which requires overpass reconstruction or peripheral closure to reduce the inflow of traffic. The "semantic+spatial" two-dimensional method based on online map in this study can identify and classify urban congestion nodes with high accuracy and low cost, and provides a replicable decision paradigm for congestion control in mega cities.