<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Emma·Z on Medium]]></title>
        <description><![CDATA[Stories by Emma·Z on Medium]]></description>
        <link>https://medium.com/@nezhalei0525?source=rss-661774fe175c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*Cz7BJrHJsx9fjgaqlfmITQ.png</url>
            <title>Stories by Emma·Z on Medium</title>
            <link>https://medium.com/@nezhalei0525?source=rss-661774fe175c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 06:46:05 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@nezhalei0525/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Basic Concepts of the Louvain Community Detection Algorithm — Node Weighted Degree]]></title>
            <link>https://medium.com/ultipa/basic-concepts-of-the-louvain-community-detection-algorithm-node-weighted-degree-199b458a8dff?source=rss-661774fe175c------2</link>
            <guid isPermaLink="false">https://medium.com/p/199b458a8dff</guid>
            <dc:creator><![CDATA[Emma·Z]]></dc:creator>
            <pubDate>Fri, 15 Aug 2025 08:15:15 GMT</pubDate>
            <atom:updated>2025-08-15T08:30:43.344Z</atom:updated>
            <content:encoded><![CDATA[<h3>Basic Concepts of the Louvain Community Detection Algorithm — Node Weighted Degree</h3><p>In real life, there are various types of social networks, such as interpersonal networks, transportation networks, and financial networks, and performing community detection on these networks is of great significance.</p><p>For example, in social networks, community detection can be applied to friend-based recommendations for products or content, calculating the influence of individuals, modeling information diffusion, identifying misinformation and bot accounts, as well as making predictions in areas such as the stock market, elections, and fraud prevention in the financial industry based on social network data.</p><p>Currently, with the explosive growth of social networks and the surging popularity of graph database applications, the Louvain algorithm stands out as one of the more complex yet particularly meaningful algorithms in the field of graph data analysis.</p><h3>1.The Historical Origins of the Louvain Algorith</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/755/1*O-nAWTtSXUSPdRf6LZGrrQ.png" /></figure><h3>2.The Basic Concept of the Louvain Algorithm: Weighted Degree</h3><p>The Louvain algorithm is a community detection algorithm based on modularity, which iteratively clusters vertices with the goal of maximizing modularity. It was proposed in 2008 by Vincent D. Blondel and colleagues at KU Leuven in Belgium. Due to its ability to efficiently produce satisfactory community detection results, it has become one of the most cited and widely used algorithms in recent years. It is mainly applied in areas such as identifying fake accounts on social networks, segmenting consumer groups and recommending products, detecting counterfeit bank cards and fraud rings, recommending banking and insurance products, and identifying corporate groups and family businesses. This section focuses specifically on weighted degree, while other topics will be covered in subsequent installments.</p><p>The basic concepts of the Louvain algorithm include:</p><ul><li>Weighted Degree</li><li>Community Aggregation</li><li>Modularity</li><li>Modularity Gain</li></ul><p>The weighted degree is the calculation of a node’s degree taking edge weights into account. The Louvain algorithm uses the concepts of node weighted degree and community weighted degree when computing modularity.</p><h4><strong>node weighted degree</strong></h4><p>Node weighted degree refers to the sum of the weights of all edges incident to a node, including edges to neighboring nodes as well as self-loops.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*sQpWlvwqWBo4YayFtiVOlQ.gif" /><figcaption>Figure 1：node weighted degree</figcaption></figure><p>As shown in Figure 1, the <strong>red node </strong>has three edges connecting to other nodes and one self-loop. Therefore, its weighted degree is 1+0.5+3+1.5=61 + 0.5 + 3 + 1.5 = 61+0.5+3+1.5=6 (note: the weight of a self-loop is counted only once).</p><h4>community weighted degree</h4><p>Community weighted degree refers to the sum of the weighted degrees of all nodes within a community.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/641/1*0OjzMvUwW9XPAVSdnE41Kw.png" /><figcaption>Figure 2: Community Weighted Degree</figcaption></figure><p>As shown in Figure 2:</p><blockquote>The<strong> red node</strong> has a weighted degree of 1+0.5+3+1.5=61 + 0.5 + 3 + 1.5 = 61+0.5+3+1.5=6</blockquote><blockquote>The<strong> green node </strong>has a weighted degree of 1+1.7=2.71 + 1.7 = 2.71+1.7=2.7</blockquote><blockquote>The <strong>blue node</strong> has a weighted degree of 0.5+0.3+2=2.80.5 + 0.3 + 2 = 2.80.5+0.3+2=2.8</blockquote><blockquote>The <strong>yellow node </strong>has a weighted degree of 333</blockquote><blockquote>Therefore, the weighted degree of Community 1 is 6+2.7+2.8+3=14.56 + 2.7 + 2.8 + 3 = 14.56+2.7+2.8+3=14.5.</blockquote><h4>internal community weighted degree</h4><p>Internal community weighted degree refers to the sum of the weights of edges whose both endpoints are within the community. In other words, it is the community’s weighted degree excluding the weights of edges connecting the community to other communities.</p><p>As shown in the figure above, <strong>Community I</strong> has three edges connecting it to the other two communities, with weights of 1.7, 0.3, and 2. Therefore, the internal weighted degree of <strong>Community I </strong>is 14.5−1.7−0.3−2=10.514.5–1.7–0.3–2 = 10.514.5−1.7−0.3−2=10.5.</p><p>Note that the internal community weighted degree is not simply the sum of the weights of edges whose endpoints are both within the community. Instead, it is calculated as <strong>twice the sum of the weights of non-self-loop edges plus the sum of the weights of self-loop edges</strong>. This is because each non-self-loop edge has two endpoints, causing the edge to be counted twice. In other words, for internal community weighted degree, self-loop edges are counted once, while all other internal edges are counted twice — since each edge connects two nodes, its weight is multiplied by 2 according to the definition.</p><p>Using the previous figure as verification, the internal weighted degree of Community I can be calculated as (1+0.5+3)×2+1.5=10.5(1 + 0.5 + 3) \times 2 + 1.5 = 10.5(1+0.5+3)×2+1.5=10.5.</p><h4>total graph weighted degree</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/933/1*QzcyEyIjqRAqNBPP7Zo6Lw.png" /><figcaption>Figure3:total graph weighted degree</figcaption></figure><p>As shown in Figure 3, the weighted degree of Community I is 14.5, Community II is 0.7×3×2+1.7=5.90.7 \times 3 \times 2 + 1.7 = 5.90.7×3×2+1.7=5.9, and Community III is 1×6×2+0.3+2=14.31 \times 6 \times 2 + 0.3 + 2 = 14.31×6×2+0.3+2=14.3. Therefore, the total graph weighted degree is 14.5+5.9+14.3=34.714.5 + 5.9 + 14.3 = 34.714.5+5.9+14.3=34.7.</p><p>If the entire graph is considered as a single community, then the total graph weighted degree can also be understood as the internal weighted degree of that community. Using the previous figure for verification, the total graph weighted degree is</p><p>(1+0.5+3+1.7+0.7×3+0.3+2+1×6)×2+1.5=34.7.(1 + 0.5 + 3 + 1.7 + 0.7 \times 3 + 0.3 + 2 + 1 \times 6) \times 2 + 1.5 = 34.7.(1+0.5+3+1.7+0.7×3+0.3+2+1×6)×2+1.5=34.7.</p><p>The results obtained from the two calculation methods can mutually verify each other and should be consistent.</p><p>Author: Wan Yi Sun</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=199b458a8dff" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ultipa/basic-concepts-of-the-louvain-community-detection-algorithm-node-weighted-degree-199b458a8dff">Basic Concepts of the Louvain Community Detection Algorithm — Node Weighted Degree</a> was originally published in <a href="https://medium.com/ultipa">Ultipa Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Graph Algorithms: The “Magic Weapon” for the Intelligent Obstacle — Avoidance of the Floor —…]]></title>
            <link>https://medium.com/ultipa/graph-algorithms-the-magic-weapon-for-the-intelligent-obstacle-avoidance-of-the-floor-e730dbb504b3?source=rss-661774fe175c------2</link>
            <guid isPermaLink="false">https://medium.com/p/e730dbb504b3</guid>
            <dc:creator><![CDATA[Emma·Z]]></dc:creator>
            <pubDate>Thu, 09 Jan 2025 02:59:20 GMT</pubDate>
            <atom:updated>2025-08-13T07:51:50.520Z</atom:updated>
            <content:encoded><![CDATA[<h3>Graph Algorithms: The “Magic Weapon” for the Intelligent Obstacle — Avoidance of the Floor — Cleaning Robot</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GGmtjWkGum5U28_H2mTL9g.png" /></figure><p>Robots are the most common smart devices in modern homes. The basic core components are the main control system (brain), sensors and other control systems (perception system), power supply system (heart), cleaning system (limbs).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JXToSVtY5V-wp1HYdhHeMA.png" /></figure><p>The path planning of sweeping robots is an important research field in robotics. In it, how to plan the optimal path during the cleaning process to ensure covering the entire area and avoiding obstacles requires the help of algorithms.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6Wc2VYeo-J_oV0WVp1E3Pg.jpeg" /></figure><p>The most common shortest path algorithms include the Dijkstra algorithm, the A * algorithm, the Bellman-ford algorithm, and the Floyd-Warshall algorithm, all of which belong to different shortest path algorithms.</p><h3><strong>Use Ultipa to model the path planning of the floor-sweeping robot：</strong></h3><p>1:Create a graph model:</p><p>· node: node can represent each location such as rooms, corridors, corners, furniture, etc. that the cleaning robot can visit.<br>· Edge:Edge is used to represent the connection between points, such as the length of the path, the difficulty of passage (e.g. through a narrow door or carpet), the type of path (straight line, curve, slope, etc.), and so on.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FlBso2r7v__udqInrKNoCA.png" /></figure><p>2.Node attributes:</p><p>· Each node can have attributes, such as the coordinates of the node, the type (starting point, end point, obstacle, etc.), whether it has been cleaned, and other status information. These attributes can all help the cleaning robot to plan and make decisions on the path.</p><p>In our daily lives, the upper-level applications are depicted in the following figure:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*E97FPsFDWKvhvMp0Z1axPw.jpeg" /></figure><p>3.Shortest path algorithm:<br>Based on the above modeling information, the cleaning robot can quickly find the optimal path from the starting point to the end point through the Dijkstra algorithm and the A * algorithm.</p><p>1)Dijkstra algorithm: In a discrete mesh of cleaning areas, the robot can calculate a path that can effectively cover the entire area based on this algorithm.</p><p>The main feature of this algorithm is that it starts from the starting point and uses the greedy algorithm strategy to traverse to the adjacent node of the closest and unvisited vertex each time until it expands to the end point.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RHZAmFdDJ1VEMUErJZFoSg.png" /></figure><p>Each number on each side of the graph represents the distance from the starting node to the terminating node (defined here as weights/weights). So, as shown in the diagram above, what is the shortest path between V1 and each point? Here’s how to use it:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fwWBPo5B4W6qxiG6gDT8YQ.png" /></figure><p>2)A * (A-Star) algorithm: The cleaning robot can dynamically plan the shortest path during the cleaning process by combining the A * algorithm with the heuristic function, and fully consider the layout of the room and the distribution of obstacles, thereby improving the efficiency of path planning.</p><p>The advantage of the A * algorithm is that it is fast. It searches for the shortest path by evaluating the distance from each node to the destination and choosing the best path. The A * algorithm uses a path evaluation formula as:</p><blockquote>f ( n ) = g ( n ) + h ( n )</blockquote><p>This also involves a problem of how the sweeping robot executes the obstacle avoidance algorithm. The sweeping robot can use a graph search-based algorithm such as A * to plan a path to avoid obstacles, ensuring that it will not collide with obstacles during the cleaning process.</p><p>As shown in the figure, let’s say we need to get from point A to the target point, and there is a wall between the two points. We rasterize the map and call the center of each square a node; there are 8 child nodes around the parent node A. This particular method reduces our search area to a 2-dimensional array. Each item of the array represents a grid, and its states are walkalbe and unwalkable. The path is found by calculating which squares we need to walk to get from point A to the target point. Once the path is found, we move from the center of one square to the center of another until we reach the destination T.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pgtL7PDucpl38mnU3qHxig.png" /></figure><p>Minimum Spanning Tree (MST) algorithms: These algorithms can be used to optimize the robot’s cleaning path, ensuring coverage of the entire area while reducing path length and improving cleaning efficiency.</p><p>The minimum spanning tree algorithm is an algorithm that attempts to select the weight and the smallest edge, so that the nodes of the whole graph are connected as much as possible. It belongs to a basic concept in graph theory and is mainly used in the solution scenarios of cost reduction and efficiency such as path optimization and seeking the lowest cost. The following are the common parameters of the minimum spanning tree algorithm.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*s8klEnkeqWz_YVstY4YMew.png" /></figure><p>In practical application scenarios, there will be dynamic changes and various complexities to consider. Graph databases can support dynamic updates of nodes and edges. This enables sweeping robots to make adjustments at any time according to changes in the real-time environment during path planning. (It should be noted that in graphs, the problem of possible loops needs to be considered. This means that we can start from a node and return to this node after going around in a circle. As long as we go around the loop once, the total weight will increase. Therefore, the path that goes around the loop cannot be the shortest path.)</p><p>Finally, the intelligence level of the sweeping robot still needs to rely on wireless sensing technology for timely observation and discovery. Among them, the application of this technology in wireless sensor networks will also use the minimum spanning tree algorithm. This algorithm models all networks such as environmental information, equipment, and routing observed in real time. Once a certain factor changes, the weight on the edge representing the graph in physical distance will also be updated in real time. Subsequently, the minimum spanning tree calculation related to edge weights will inevitably be adjusted in real time to finally ensure that the sweeping robot can easily deal with obstacle blockages or various emergencies during the cleaning process.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/977/1*De6w7p2PHr0G-zZmgQsisw.jpeg" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e730dbb504b3" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ultipa/graph-algorithms-the-magic-weapon-for-the-intelligent-obstacle-avoidance-of-the-floor-e730dbb504b3">Graph Algorithms: The “Magic Weapon” for the Intelligent Obstacle — Avoidance of the Floor —…</a> was originally published in <a href="https://medium.com/ultipa">Ultipa Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[From data to intelligence, decrypting the “intelligent brain” of the logistics industry — the…]]></title>
            <link>https://medium.com/ultipa/from-data-to-intelligence-decrypting-the-intelligent-brain-of-the-logistics-industry-the-8a0e1dc9f629?source=rss-661774fe175c------2</link>
            <guid isPermaLink="false">https://medium.com/p/8a0e1dc9f629</guid>
            <dc:creator><![CDATA[Emma·Z]]></dc:creator>
            <pubDate>Tue, 31 Dec 2024 03:47:06 GMT</pubDate>
            <atom:updated>2025-08-13T07:52:13.795Z</atom:updated>
            <content:encoded><![CDATA[<h3>From data to intelligence, decrypting the “intelligent brain” of the logistics industry — the disruptive power of graph databases</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KOiNdxkm8H9q4tL9hAjjDg.png" /></figure><p>As early as 3000 years ago, ancient Egypt, Greece, and Rome began to use shipping systems to start the transportation and distribution of materials, and formed ancient super logistics centers such as Ostia and Alexandria. In ancient China and ancient Inca, the perfect post station system provided upgraded support in terms of materials, personnel, and supplies, and guaranteed the speed and efficiency of long-distance transportation in terms of systems.</p><p>Entering the 21st century, in the modern logistics world, the pursuit of speed and efficiency is still the key, whether it is from the factory to the warehouse, or from the warehouse to the customer, every second is crucial. For example, a super e-commerce platform like Amazon, whose logistics network covers the world, involving thousands of warehouses, tens of thousands of transportation routes, and millions and tens of millions of customer orders… So behind the huge and complex logistics network, a powerful system is needed to manage and optimize all processes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*iGvV26uaYUN1MVSz.jpeg" /></figure><p>In particular, with the increasing complexity and scale of logistics services, the processing power of traditional relational databases in the past has obviously reached the ceiling. Due to the fact that their data is stored in tables, not only has the amount of computation been multiplied, but the relationship between the data is also hidden in the vast number of queries. This is the natural advantage of the Graph Database. Its core lies in the processing of various relationships between the data, which is especially suitable for complex logistics networks.</p><p>In the graph database, the data is represented as node, and the relationship between the data is represented as edges. This data structure is like a huge network diagram. Points represent various entities in logistics, such as warehouses, goods, transportation vehicles, etc., and edges represent the relationships between these entities, such as transportation routes, transportation times, transportation costs, etc.</p><p>And in the logistics network optimization, the graph database can more accurately describe the relationship between logistics points, providing more accurate data support for the optimization algorithm!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QFxq6D0M-u-M0DLojivfMw.png" /></figure><p>And in the logistics network optimization, the graph database can more accurately describe the relationship between logistics points, providing more accurate data support for the optimization algorithm!</p><ol><li>Optimal route planning</li></ol><p>In logistics, how can we deliver goods to their destinations at the lowest cost and in the shortest possible time?</p><p>Suppose you are a delivery person who needs to deliver goods from a warehouse to multiple locations. This seemingly simple task actually involves numerous route choices. Then, how to determine an optimal transportation route is the key to improving efficiency and reducing costs.</p><p>Using a graph database, a logistics network can be modelled as a graph, with points representing distribution centres or customers and edges representing transportation routes, and each edge given weights such as distance or time. By running the Dijkstra algorithm or the A * algorithm, the graph database can find the optimal path for you in a short time — like having a smart brain.</p><pre>ab().src({name == &quot;Warehouse_A&quot;}).dest({name == &quot;Customer_B&quot;}).depth(10).shortest(@connection.distance) as p<br>return p</pre><p>Explanation: Use the Ulitpa GQL-UQL query to search for all possible routes between “Warehouse_A” and “Customer_B”, and calculate the total cost (based on distance) for each route. Sort the query results by cost and return the route with the lowest cost to assist logistics in planning the optimal route.</p><p>2.Transparency in the supply chain</p><p>The coffee we use every day to refresh our minds has gone through multiple links from plantations to processing plants, to packaging, transportation, shopping malls (stores), etc. We can use a graph database to connect these links to form a complete supply chain diagram.</p><p>In the graph database, the various links of the supply chain can be modeled as nodes, and the relationships in the supply chain (such as supply and distribution) can be modeled as edges. Through Yingtu GQL, various types of information in the supply chain can be queried in real time to analyze potential bottlenecks and risks. For example, there will be multiple plantations, suppliers, distributors, retailers, etc., and the relationship is complex. If there is a problem in a certain link (such as shortage of raw materials), the system will immediately flag this point and remind managers to take countermeasures at the first time.</p><pre>n({@Supplier || @Manufacturer || @Distributor}).e()[:5].n({@Supplier || @Manufacturer || @Distributor}) as p return p{*}</pre><p>Explanation: The query returns the relationship between suppliers, manufacturers, and distributors, helping managers understand the overall structure of the supply chain and optimize supply chain management strategies accordingly.</p><p>3.Real-time tracking and response</p><p>A key challenge in the logistics industry is to monitor the transportation status of goods in real time, especially in the context of global distribution and multimodal transportation (multimodal transportation), where graph databases play a huge role.</p><p>Graph databases can dynamically update the status of each node and edge (that is, each piece of cargo) in the logistics network and accurately mark their positions within the entire logistics network. Logistics managers can display the real-time status through UQL queries and check the current position of the cargo and predict the arrival time at the earliest opportunity.</p><pre>find().nodes({@Cargo.name == &#39;CARGO4530&#39;}) as n<br>return n{*}</pre><p>Explanation: Query returns the current location information and status of a specific shipment “CARGO4530”, helping managers monitor the progress of cargo transportation in real time.</p><p>4.Risk prediction and contingency plan</p><p>In the logistics business, potential risks are everywhere, such as bad weather, natural disasters, traffic accidents, human factors, etc., as well as abnormal behaviors in the logistics system, such as delays, misdeliveries, fraud, etc., which are often difficult to detect and may cause chain interruptions at any time, affecting the delivery of goods on time. Yingtu Database monitors the abnormal dynamics of each link at any time through pattern analysis of a large amount of data, keenly detects abnormalities, and arranges emergency plans in advance. For example, if the time of a certain transportation line is suddenly extended, the graph database will flag it as a potential problem and remind managers to investigate further.</p><p>The Ultipa database can support in-depth searches, queries, and calculations of more than 30 layers in complex graph sets with hundreds of billions of data volumes. It can achieve ultra-deep real-time penetration for datasets of any magnitude, quickly gaining insights into the deep, subtle, and complex correlations hidden in the data, and rapidly locking down risk points to enable early detection, early preparation, and early prevention.</p><p>Above we have an intuitive and perceptual understanding of the application of graph technology in logistics management. But there are still people who ask, we can also achieve it with relational databases or other database systems, why do we have to use graph databases? Yes, other databases are certainly acceptable, but for complex application scenarios such as logistics, graph technology has natural advantages. Its data structure itself is a map of the real world, capable of efficiently managing and processing a large number of complex relationships involving multiple layers and dimensions, including warehouses, distribution centers, transportation routes, suppliers, etc., providing a more efficient solution than traditional and other databases.</p><p>What exactly is an algorithm? In simple terms, it is a specific calculation method and processing step of limited length taken to solve a certain problem. That is to say, there are two types of outputs of an algorithm. The first is the result produced by the algorithm (clustering, classification, predicted value, etc.), and the second is the rule produced by the algorithm. The Yingtu database has the richest high-performance and scalable graph algorithm package in the international scope — more than 100 kinds. By supporting a flexible Demi-schema hybrid mode, it can truly empower logistics enterprise managers to deeply query data in various links, and support real-time query and dynamic update. It can maintain efficient performance and scalability in a large-scale logistics network, enabling logistics companies to respond to business needs and market changes in real time.</p><p>What exactly is an algorithm? Briefly speaking, it refers to specific computational methods and processing steps of a limited length adopted to solve a certain problem. That is to say, there are two kinds of outputs of an algorithm. The first is the results produced by the algorithm (such as clustering, classification, predicted values, etc.), and the second is the rules produced by the algorithm. The Ultipa graph database has the richest high-performance and scalable graph algorithm package in the international arena — with more than 100 kinds. Moreover, by supporting the flexible demi-schema mixed mode, it can truly empower logistics enterprise managers to conduct in-depth queries on the data of various links. It also supports real-time queries and dynamic updates, and can maintain high performance and scalability in large-scale logistics networks, enabling logistics companies to respond to business needs and market changes in real time.</p><p>Not all graph databases are the same. The so-called donkey dung egg surface light, we have seen too many low-level engines do very badly, but the surface effort (various visual charts) has made very prosperous manufacturer products. For example, some graph databases are non-native RDF architectures, which can only do some shallow processing, or the data needs to be transported a lot, and it needs to be placed in memory to do distributed computing, or it can only support static data, without a lot of persistence layers… These are not really able to give the upper-level business the implementation and value of the application!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8a0e1dc9f629" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ultipa/from-data-to-intelligence-decrypting-the-intelligent-brain-of-the-logistics-industry-the-8a0e1dc9f629">From data to intelligence, decrypting the “intelligent brain” of the logistics industry — the…</a> was originally published in <a href="https://medium.com/ultipa">Ultipa Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The power station of the “Graph” world]]></title>
            <link>https://medium.com/ultipa/the-power-station-of-the-graph-world-2de3456188b5?source=rss-661774fe175c------2</link>
            <guid isPermaLink="false">https://medium.com/p/2de3456188b5</guid>
            <dc:creator><![CDATA[Emma·Z]]></dc:creator>
            <pubDate>Thu, 12 Dec 2024 06:33:53 GMT</pubDate>
            <atom:updated>2025-08-13T07:49:32.496Z</atom:updated>
            <content:encoded><![CDATA[<p>Ultipa Powerhouse, literally translated as power station, power station, is a milestone for Ultipa since the release of the high-density parallel graph computing engine in 2019, including a comprehensive innovation of the entire product architecture, providing fast and powerful power support for computing, analyzing and storing large amounts of complex data. Ultipa users are currently using it to process hundreds of millions of transactions per day in real time or for scientific research.</p><h3>Graph database + data visualization</h3><p>Ultipa’s products mainly have two cores: the bottom-level graph database and the upper-level data analytics and visualization software.</p><p>Ultipa’s products mainly have two cores: the underlying graph database and the upper-level data analytics and visualization software. RealGraph Database is a distributed, scalable, and elastic high-performance native graph database that can host trillion-level point-edge datasets. It provides real-time microsecond and millisecond-level query and analysis through a hyperconverged unified graph computing architecture. Since Ultipa released the industry’s first high-density parallel graph computing engine in 2019, it has undergone many iterations:</p><blockquote>1、In 2019, Ultipa released a high-density parallel graph computing engine, with the main performance indicators benchmarking the industry’s leading products and providing exponential performance advantages.<br>2、In 2020, the Ultipa graph database will be released and industry internal testing will begin.<br>3、In 2021, Ultipa released the first HTAP graph engine to support distributed clusters, beginning large-scale commercial deployment — and supporting multiple challenging and innovative business scenarios and killer applications in the industry, such as Oracle cash flow engine based on real-time graph computing white box cracking, real-time liquidity risk management and attribution analysis, cross-metric attribution analysis, and more.<br>4、In 2022, Ultipa will release cloud-native SaaS/DBaaS product tools, plug-and-play graph algorithms, one-click deployment, and more killer applications.<br>5、In 2022, Ultipa will release the Blaze version, pioneering a half-mode blended mode design for maximum flexibility and more advanced, in-depth queries with intuitive data modeling.<br>6. 2023 Ultipa Blaze enterprise-level rapid iteration.<br>7. Ultipa Powerhouse will be released in 2024.</blockquote><p>It is understood that at the beginning of the development of Ultipa products, researchers found that many so-called graph platforms on the market, such as “so-and-so smart city” and “so-and-so smart platform”, were just displays of big data screens, lacked the ability to self-develop the underlying technical architecture, and the graph storage and graph computing engines were either non-native, multi-modal, or based on Neo4j Community Edition or open-source Janus graph and Arango DB for shell transformation, or even based on traditional relational databases or big data frameworks such as Apache Spark. Therefore, the Ultipa team started from scratch with kernel development and architecture innovation, and developed a complete set of truly computing power, intelligence, and wisdom graph systems that combine the underlying technology high-density, high-concurrency hardware-accelerated dynamic graph engine with the upper-level application Knowledge Graph platform.</p><h3>What makes Ultipa Power house unique?</h3><p>Ultipa Powerhouse is based on C++ implementation and adopts a new hyper-converged architecture. It consists of four main parts: Shard server, HDC server, Name Server server and Meta server. It has high-performance graph computing power and elastic scalability.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*s0cnziOx8FxDV22wJJegjg.png" /></figure><p>One is the unified architecture, which further enhances horizontally scalable memory-computing integrated clusters (SCC) and high-density computing (HDC), as well as a unified architecture design with low memory consumption. Graph processing systems often face a trade-off between data locality and computational efficiency. Traditional distributed computing clusters, while supporting horizontal scaling, struggle to maintain optimal data locality due to their general design. On the other hand, high-density computing nodes have the ability to drill down more efficiently, but they also lack the required scalability and flexibility when faced with diverse elastic graph workloads.</p><p>Ultipa Powerhouse bridges this gap by seamlessly integrating scalability and optimal performance. It ensures efficient graph traversal and computation while meeting enterprise-scale large-scale graph requirements (e.g. hundreds of billions, trillions) and providing the flexibility required for elastic, dynamic workloads.</p><p>At this year’s SIGMOD 2024 conference, an international academic summit conference on global databases, Ultipa gave a speech entitled “Unified Graph Framework with SCC (Storage-Computing Coupling) and HDC (High Density Computing) Clusters”, and presented a paper of the same name at the conference. It is understood that the SIGMOD conference is initiated by the American Association for Computing Machinery (ACM) Professional Committee on Data Management (SIGMOD) and is one of the most influential international academic conferences in the fields of data management, data processing and data storage.</p><p>In the application transformation of theory and practice, the paper submitted by the Ultipa team at the ACM SIGMOD 2023 conference is “Designing Highly Scalable Graph Database Systems without Exponential Performance Degradation”, which clarifies the architecture design of Ultipa Powerhouse: real-time deep drilling capability of HTAP; “artificial intelligence sharding” of grid architecture with Name Server and Meta Server; horizontal automatic sharding architecture with NS Server, Meta Server and Shard Server; and the integration of HTAP and automatic sharding architecture to achieve the compatibility of deep computing and mass storage. It is worth mentioning that Ultipa conducts comprehensive planning from the perspective of architectural evolution, from HTAP distributed architecture to Grid grid to automatic sharding architecture. In particular, the sharding architecture illustrates the diversity and complexity involved in designing complex and capable graph systems, which includes comprehensive performance demonstrations including overall performance, reliability, and green performance.</p><p>The second is high performance. For computing-first graph databases, whether high-density concurrency can be achieved in all aspects of the underlying kernel, data structure, super node penetration, engineering work of algo, etc., so that the release of hardware computing power and the full acceleration of software are the core elements that determine whether a graph database has high performance and is capable of complex computing tasks in various scenarios. Ultipa’s technological innovation is not limited to the original design of the technical core. In October 2023, Ultipa won the world’s first patent for “super node penetration” — one of the key technologies that enable real-time or near-real-time deep drilling and complex computing. Ultipa is the only dynamic graph algorithm technology at home and abroad that realizes the ability of “write-and-calculate” and “full calculation”, which not only takes into account the calculation efficiency but also ensures the accuracy of the calculation.</p><p>Third, it is the first to release new international standards that support GQL. Ultipa launched the Powerhouse Beta version that supports GQL in September this year, and it is also the first graph manufacturer in the industry to support GQL.</p><p>Ultipa has been dedicated to building a world-class graph database system since its inception. It is worth mentioning that, as a world-class authoritative graph vendor, Neo4j’s JAVA programming language-centric architecture and some core design flaws, such as the inability to handle super nodes and not allowing real-time data processing, have been the bottleneck of its later technology development. Ultipa has comprehensively summarized its experience and lessons, and has carried out a comprehensive and in-depth exploration of its own technical research and innovation. In practical application performance comparisons, such as in regulatory compliance and audit scenarios, Neo4j cannot penetrate 5-hop and below data in real time when dealing with drill-down penetration based on the full industrial and commercial enterprise graph, while Yingtu has achieved 32-hop deep penetration and runs 10,000 times faster than Neo4j — which means a leap from T + 1 to real-time capabilities.</p><h3>Comprehensive toolchain: providing a full journey “map” experience</h3><p>Ultipa has built a one-stop graph computing tool that covers the full journey “graph” experience from configuration deployment, data scheduling, transport modeling, visual analysis queries, programming runs, and business services. Key tools include:z</p><blockquote>▪ Ultipa-Manager: A full stack highly visual, intuitive, highly scalable, low-code graph database management system.<br>▪ Ultipa-Transporter: An ultra-high-performance data import/export tool that supports multiple data sources and data formats.<br>▪ Ultipa-Drivers: A mature API/SDK that seamlessly supports application development, with support for Java, Go, Python, NodeJS, C #, and RESTful APIs.<br>▪ Ultipa-CLI: A cross-platform command-line interface (CLI) that can execute GQL and UQL queries in a variety of environments.<br>▪ GQL: an ISO-compliant graph query language that sets global standards.<br>▪ Ultipa-GQL: Ultipa query language, designed for the low-level performance of graph databases.<br>▪ Ultipa graph algorithm package: It has the world’s richest graph algorithm set, more than 100 algorithms, and can be hot-swapped and extended through the EXTA interface to help data science applications.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zFRC71hLy3bLvPg8FiWsNw.png" /></figure><p>The end-to-end product operation experience has always been the focus of Ultipa’s research and development. For example, high-speed loading of dynamic data, dynamic loading of graph algorithms, algorithm scalability and real-time, graph data lake visualization, etc. The Yingtu team focuses on empowering customers and partners to achieve cost reduction and efficiency and enhance the user experience in real-world scenarios and even extreme application scenarios.</p><h3>Augmented Intelligence Revolution: Driving digital transformation for organizations</h3><p>Today, we can clearly see that artificial intelligence and big data infrastructure are converging, people’s work lives are being deeply reshaped, and Ultipa Graph XAI graph enhancement technology is essential to improve the accuracy, transparency and explainability of AI results, becoming an inevitable technical path to realize digital and intelligent transformation.</p><p>In recent years, Ultipa has successfully handled some of the most complex financial data scenarios at home and abroad, pioneering the application of graph-augmented intelligence technologies, including but not limited to operational analysis, liquidity risk, intelligent auditing, transaction interception, fund flow, regulatory compliance, and more.</p><p>In the international competition, Ultipa’s graph calculation engine based on real-time graph enhanced intelligence is more than 10–100 times more efficient than Silicon Valley congeneric products in terms of calculation timeliness, and more than 1000 times faster than Oracle database when measured by banking risk management indicators. In the process of building intelligent management and asset liability management charts, it successfully cracked the two “black boxes” of Oracle cash flow engine and rule engine. It is a technology that transforms the core scenarios of the financial industry in the way of XAI graph enhanced intelligence. It realizes the performance characteristics of “fast calculation”, “accurate calculation”, “accurate”, “deep calculation”, “real-time”, “visualization”, “explainable, white box, and verifiable” at the practical level. Furthermore, the existing AI, ML, and LLM large-scale model architectures feature model efficiency, acceleration, and greatly improved prediction accuracy and interpretability, making them disruptive innovations and breakthroughs on a global scale.</p><p>In 2024, Ultipa was selected into the UK FinTech Global “DataTech50 for 2024” list and the ESG FinTech100 list; last year, as a representative manufacturer of China’s graph database, it competed with S &amp; PGlobal, E &amp; Y and other manufacturers on the same stage, and finally won the 24th British Bank Technology Award — the first place in the Future Technology (artificial intelligence and data) Award and the finals of the China Shenzhen International Financial Technology Competition. It competed fiercely with 103 projects along the way, and finally won the first prize of the “artificial intelligence and big data” track and the second place in the finals as the only company to enter the top three with AI software.</p><p>Just earlier this November, a research report from international analytics firm Bloor Research gave a comprehensive look at how Ultipa Powerhouse empowers the world in terms of performance, scalability, and built-in artificial intelligence!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/926/1*X_DIYhXqU6Jg5f_Kw49uig.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2de3456188b5" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ultipa/the-power-station-of-the-graph-world-2de3456188b5">The power station of the “Graph” world</a> was originally published in <a href="https://medium.com/ultipa">Ultipa Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>