Optimizing Chinese Train Transfers with Graph Databases: A Step‑by‑Step Guide

Background

During the National Day holiday many direct train tickets are sold out; 12306 offers limited transfer options, often requiring long layovers. Train stations and schedules naturally form a graph, so using a graph database can find optimal transfer routes.

Data Preparation

Train schedule data (e.g., 列车时刻表.txt ) is downloaded from public sources. The raw files are parsed in Python to extract stations, stops, and trains, then written to CSV files for vertices ( station.csv, train.csv, stop.csv) and edges ( next.csv, arrive_at.csv).

mkdir -p graph_data/edge
mkdir -p graph_data/vertex

def read_files(path, skip_comment=False, skip_header=False):
    space_read = True if not skip_comment else False
    header_read = True if not skip_header else False
    with open(path, 'r') as f:
        for line in f.readlines():
            if not header_read:
                header_read = True
                continue
            if not space_read:
                if line == "
":
                    space_read = True
                continue
            yield line

Schema Definition

The graph model defines three node types: Station , Stop (a train’s arrival at a station) and Train . Edges NEXT connect consecutive stops of the same train, and ARRIVE_AT link stops to stations.

Importing into GES

The CSV files are uploaded to OBS, a GES instance is created (choose at least a million‑edge specification), and the data is imported via the GES console or API.

Querying Transfer Routes

Cypher queries can retrieve direct trains, multi‑stop routes, and transfer options. Example: find trains from Nanjing South to Taiyuan with a single transfer, enforce a minimum 15‑minute layover, and limit results.

match (n:Station)<-[:ARRIVE_AT]-(s:Stop) where id(n) in ['南京南']
match p=(s)-[:NEXT*1..30]->(s1) where s1.station contains '太原'
return s.trainNo as `车次`,
       subString(toString(s.arrives),11) as `出发`,
       subString(toString(s1.departs),11) as `到达`,
       subString(toString(datetime(timestamp(s1.departs)-timestamp(s.arrives))),11) as `耗时`,
       [x in nodes(p)|x.station] as `途径`
limit 10

Additional queries filter by departure time (e.g., before 11 am) or restrict transfers to specific stations such as 德州.

Conclusion

Compared with the 12306 website, graph‑database queries provide more transfer alternatives and shorter travel times, enabling travelers to select the best itinerary based on ticket availability and personal preferences.