April 2021

Graphing POW Camp Mistreatment in Wartime Japan

This graph database is a pilot project to help illuminate the connections and relationships among seemingly discrete war crimes trials that the Allied authorities held in Asia and the Pacific following the end of World War II. More than 2,240 trials took place involving some 5,700 suspected war criminals as accused at 51 separate locations in the former theaters of war. This pilot focuses on 235 cases concerning the Japanese mistreatment of Allied prisoners of war in wartime Japan, as had been made at the U.S. Army’s military commission at Yokohama (1946-1949).

A Guide to Using the Graph Database

This project uses the POLE model as a framework for modeling the crimes detailed in the U.S. Yokohama trials. For more about the POLE Model, see the WCDI Data and Visualization Exhibit.

Sources

The record of an Allied war crimes trial commonly contains five sets of legal and factual findings, namely, those of (1) the prosecution, (2) the defense, (3) the judges, (4) the reviewing authority, and (5) the confirming authority.1 Instead of attempting to analyze five sets of diverse and often mutually conflicting legal and factual findings exhaustively, this pilot has limited its task to analyzing the prosecution’s charges and specifications, as summarized in reviews of individual U.S. Yokohama trials, so as to extract the data as a set of essential POLE data to ingest into a graph database.

Reviews of U.S. Yokohama trials in a digitized version can be accessed at the website of the International Research and Documentation Centre for War Crimes Trials (ICWC) at Philipps University Marburg.

Example of the first page of a typical review Example of a summary of charges, specifications, and findings in a typical review

The review of each trial commonly contains brief biographical information of the accused, a summary of charges, specifications and verdicts, a detailed summary of the prosecution’s case, a detailed summary of the defense case, recommendations by the reviewing authority, and the decision of the confirming authority. It should be noted that the reviews are no substitute for the actual records of trials (i.e. official records of court proceedings such as the charge sheet, the bill of particulars, transcripts of court proceedings, court exhibits, and the court’s decision), which are deposited at the National Archives and Records Administration (NARA) at College Park, MD.2

POLE data, statistics, color coding, and nodes & relationships

Specific data that this project sought are as follows:

  • Case number of individual trial
  • Name of accused
  • Prisoner-of-war camp affiliation of accused
  • Superior-subordinate relations among accused (if indicated in specifications)
  • Crime type3
  • Crime date
  • Name of victim if indicated in specifications
  • Number assigned to each charge and specification
  • Mode of liability4

The total number of POLE data points collected for the pilot project is as follows:

Nodes Count (10,835)
Trials 235
Episodes of crimes / charges and specifications 6,224
Locations / prisoner-of-war camps 121
Members of the Japanese military (mostly defendants at the U.S. Yokohama trials) 616
Victims of war crimes (Allied prisoners of war) 3,639
Relationships (superior-subordinate relationships and affiliations) Count (20,945)

For visualization purposes the nodes are color coded to make it easy for users to distinguish between the nodes representing the trials (blue), episodes of crimes / charges and specifications (orange), locations/prisoner-of-war camps (beige), and persons / accused and victims (red). These colors are customizable, and the default colors will likely vary across different machines. Specific types of relationships that connect the nodes – i.e. superior-subordinate relations and affiliations – are also recorded and labeled in graph visualizations.

Schema showing all possible nodes and relationships in the database Example of graph with nodes and relationships labeled by type Example highlighting subordinate/superior relationship Example highlighting affiliation between a trial defendant and a location Example highlighting multiple relationships of a defendant

Users can explore additional nodes and relationships by clicking a graph icon that accompanies each node. Clicking the same icon for the second time will hide the expanded nodes and relationships. The lock icon will lock the node in place in the visualization, while the eye icon will remove the node (and all relationships between the node and other nodes) from the visualization.

Option wheel for locking, hiding, or expanding a node Graph showing relationship between locations Detail showing relationship between a trial defendant, multiple charges, and a location

Additional POLE data that are recorded as attributes in each node. One can have them revealed on the face of the node by clicking first the relevant color-coded category icon at the top toolbar, and then select the desired data classification at the bottom toolbar. The text shown on the node changes according to the selection made on the toolbars.

Neo4j graph with visualization options Neo4j visualization options: top Neo4j visualization options: bottom Multiple label options for nodes - mode of liability Multiple label options for nodes - number of victims Multiple label options for nodes - crime type Multiple label options for nodes - case number and charge numbers

Human errors in data analysis & extraction, and ambiguities in texts

Unlike the digital world of the present, the records of historical war crimes trials are commonly unprocessed and, even if available in digital format, rarely text-searchable or otherwise unfit for direct application with the current automated and digital tools. Consequently, one needs to resort to manual workflows to extract and analyze the sources and generate an electronic data set. This process is necessarily prone to create errors even if the data are handled by meticulous and detailed-oriented researchers. Human errors may be minor kinds on their face – such as the mistaken use of spacing, commas, semi-colons, etc. – but they could cause unnecessary duplication of nodes and relationships or other confounding outcomes when ingested into the graph databases. This pilot project has corrected errors by proofreading the datasets repeatedly. There likely remains undetected errors, however, which will be corrected as they are discovered.

It is not uncommon for historical sources to contain ambiguities, especially when the sources consist largely of texts and narratives and not structured or coded data per se. For instance, an episode of war crimes as articulated in a charge sheet might be fit to be classified as not one but two or more “crime types.” Similarly, some charges and specifications are constructed in a multilayered manner to leave some levels of ambiguities as to which specific mode of liability should apply to an accused. Such instances of ambiguities required us to make a judgment call from time to time, in order that we would generate a less-ambiguous dataset that the graph databases could ingest. Those who are interested in the textual dimension of the trials should peruse the reviews and, ultimately, the actual records of the trials themselves.

Making queries and exploring the graph

A researcher would typically think of a question they would like to explore, then construct the question using a graph query language. In this case, we use Cypher, Neo4j’s query language.

A Cypher query

Until an online interactive system or sandbox is publicly accessible, we provide several sample queries as static displays on how one might explore the data.

  1. Graph query showing all of the 235 U.S. Yokohama trials that focused on the prosecution of suspected war criminals for the mistreatment of Allied prisoners of war in wartime Japan

    The first image shows the POW camp cases as discrete events (represented as blue Event nodes) that are unrelated to one another. The beige Location nodes represent seven prisoner-of-war camp groups in Japan (Hakodate, Sendai, Tokyo, Nagoya, Osaka, Hiroshima, and Fukuoka).

    Graph of disconnected nodes

    In the next image, the node representing the Fukuoka Area POW Camp Group is expanded so that one can see the affiliated prisoner-of-war camps.

    Connected nodes within a group of disconnected nodes Detail showing relationships between locations in the Fukuoka Area POW Camp Group

    Clicking one of the beige Location nodes representing affiliated prisoner-of-war camps will display People nodes in red with the names of Japanese individuals who were tried at the U.S. Yokohama trials in connection with that particular location.

    Detail showing defendants affiliated with Fukuoka Branch Camp No. 2

    Clicking one of the People nodes in red representing a suspected war criminal will display Event nodes of the relevant trial and episodes of crimes/ charges and specifications.

    Detail showing nodes connected to defendant Akiyama, Fukujiro

    Clicking one of the Event nodes here in orange representing an episode of crime / charge and specification will display the names of victims if they have been indicated in the summary of charges and specifications in the review.

    Detail showing additional relationships within Trial 177

    By repeating this process, one can begin to see relationships among the trials that otherwise may not be apparent to researchers.

    Detail showing relationship between multiple trials involving Fukuoka Branch Camp No. 2 Detail showing all trials of defendants affiliated with Fukuoka Branch Camp No. 2 Detail showing all trials in the Fukuoka Area POW Camp Group
  2. Graph query of the joint trial of SUGASAWA Ijū, formerly a colonel and commander of the Fukuoka Area POW Camp Group, and three others (Case Number 224)

    The first image shows the result of a query on SUGASAWA Ijū, formerly commander of the Fukuoka Area POW Camp Group in Kyūshū in southern Japan.

    Graph of trial of Sugasawa, Iju

    Pulling the nodes away from one another and reorganizing them enables one to see more clearly the relationships among these varied-colored nodes representing persons, charges and specifications, a prisoner-of-war camp group, and a trial.

    Graph of trial of Sugasawa, Iju, highlighting relationships between people

    By clicking an Event node representing trial, charges and specifications will display the nodes of victims insofar as they are indicated in the summary of charges and specifications in the review.

    Graph of trial of Sugasawa, Iju, demonstrating use of the option wheel Detail of victim mentioned in multiple charges in the trial of Sugasawa, Iju Detail showing multiple victims in the trial of Sugasawa, Iju

    Clicking a name node of an accused will bring out other critical data about that person in the forms of nodes and relationships. In this instance, a new node that represents an additional location is exposed. Clicking this node reveals its relationship with the other prisoner-of-war camp to which Sugasawa had belonged.

    Detail showing the option wheel Detail showing the option wheel Detail showing result of expanding a node via the option wheel

    This process may be repeated to explore the connection between the Sugasawa Trial with other cases.

    Detail of the option wheel Detail showing connections between Fukuoka area camps Detail of Fukuoka area camps, with Fukuoka Branch Camp No. 1 separated from others Detail of the option wheel using Fukuoka Branch Camp No. 1 as an example Graph showing defendants affiliated with Fukuoka Branch Camp No. 1 Detail of the option wheel Graph showing a single trial associated with defendants affiliated with Fukuoka Branch Camp No. 1 Graph showing multiple trials affiliated with Fukuoka Branch Camp No. 1 Graph showing trials associated with defendants affiliated with Fukuoka Branch Camp No. 1 Graph showing relationship between the trial of Sugasawa, Iju and defendants affiliated with Fukuoka Branch Camp No. 1
  3. Graph query of the joint trial of KOJIMA Ichisaku, former CEO of Niigata Sea and Land Transportation Company, and six others (Case Number 314)

    This is one of a few U.S. Yokohama cases where individuals in the business sector were prosecuted as suspected war criminals. The chief accused is KOJIMA Ichisaku, former CEO of the Niigata Sea and Land Transportation Company, and the rest of the accused were company employees. This image show how a graph database looks like when one makes a query on Kojima.

    Graph showing relationships in Trial 314

    Pulling the nodes away from one another and reorganizing them enables one to see more clearly the relationships among these varied-colored nodes representing persons, charges, and a prisoner-of-war camp. It may be noted that, while Kojima was the company CEO, this graph does not indicate any superior-subordinate relationship between him and the rest of the accused. This is because our dataset for this pilot project did not record superior-subordinate relationships unless the summary of charges and specifications in a review expressly recognized them. One can learn that Kojima was a company CEO, however, by reading the rest of the case’s review where details of the prosecution’s case, the defense contention, outcomes of the trial, and recommendations by the reviewing authority and the confirming authorities are shown.

    Graph showing relationship between Trial 314, defendants, and Tokyo Branch Camp No. 5 at Niigata

    Clicking the node for the prisoner-of-war camp will expose other individuals who had been prosecuted at the U.S. Yokohama trials in connection with war crimes committed at the same prisoner-of-war camp.

    Option wheel centered on Tokyo Branch Camp No. 5 at Niigata Graph showing nodes connected to Tokyo Branch Camp No. 5 at Niigata

    Repeating the foregoing exercise will eventually lead one to come across the names of Colonel SUZUKI Kunji and Colonel SAKABA Kaname, who served as commander of the Tokyo Area POW Camp Group successively (1942-1944, and 1944-1945, respectively). See the review of the joint trial of SUZUKI Kunji and SAKABA Kaname (Case Number 305).

    Option wheel centered on the Tokyo Area POW Camp Group Graph showing locations associated with the Tokyo Area POW Camp Group Graph focused on Suzuki, Kunji, commander of the Tokyo Area POW Camp Group Graph focused on Suzuki, Kunji

    Clicking the node representing Suzuki shows that there are a number of his subordinates who had been prosecuted at the U.S. Yokohama trials.

    Graph showing subordinates of Suzuki, Kunji
  4. Graph query of the trial of HATAKAYAMA [Hatakeyama] Toshio, formerly a colonel and commander of the Hakodate Area POW Camp Group (Case Number 203)

    The first image shows what the graph looks like when one makes a query on HATAKEYAMA [shown here as “HATAKAYAMA”] Toshio, formerly commander of the Hakodate Area POW Camp Group in Hokkaidō in northern Japan. This was the trial of a single individual but this query allows one to see a number of Hatakeyama’s former subordinates who had also been prosecuted at the U.S. Yokohama trials.

    Graph showing relationships in Trial 203 Graph centered on Trial 203 with nodes divided by type Graph centered on Trial 203 with nodes divided by type

    Clicking the node of one of the subordinates will expose other nodes and relationships that otherwise may be unknown to researchers.

    Detail of the option wheel Graph showing relationship between Trial 8 and Trial 203 Detail of the option wheel Detail showing victim mentioned in both Trial 203 and Trial 8 Detail of the option wheel Detail of charges in Trial 4 Detail of the option wheel Detail showing relationships between defendants and locations in the Hakodate Area POW Camp Group Detail of the option wheel Detail of defendants affiliated with Hakodate Main Camp Detail of specific defendants affiliated with Hakodate Main Camp Detail showing trials of defendants affiliated with Hakodate Main Camp

About This Graph Database

David Gustavsen was the chief developer of this graph database, assisted by Peter Bushell, Lauren Hauck, and John Winnicki who analyzed the U.S. Yokohama trials’ reviews and created a dataset for this project.


  1. The Allied war crimes trials in neither Europe nor the Far East had any appeals courts but instead the reviewing and confirming authorities that reviewed, made recommendations, and, in the case of the U.S. Yokohama trials, “approve, mitigate, remit in whole or in part, commute, suspend, reduce or otherwise alter the sentence imposed, or (without prejudice to accused) remand the case for rehearing before a new military commission.” Article 5 (h) of “Regulations Governing the Trials of Accused War Criminals” (December 5, 1945), also known as “SCAP Regulations,” which applied to the trials by the U.S. Army’s military commission at Yokohama.
  2. The Virtual Tribunal Initiative at the Center for Human Rights and International Justice at Stanford University, which the Center leads in collaboration with the Stanford University Libraries, is seeking to develop a comprehensive text-searchable database of international criminal trials since the time of the Nuremberg and Tokyo Trials (1945-1949) to the contemporary ones. While the project is under way, those who wish to take a deeper look at individual U.S. war crimes trials should visit NARA to view the paper copy of the trial records in person. A comprehensive guide to archival records on Japanese war crimes, which is available online, allows researchers to identify in advance the relevant record groups when planning an archival trip.
  3. A set of crime-type “codes” was devised for analyzing and extracting data from descriptive narratives in historical documents – in this instance, reviews in the U.S. Yokohama Trials – so that a standardized dataset could be generated. This set of crime-type codes uses as a model a list of crime types in Appendix D, “Particulars of Breaches,” in the IMTFE indictment.
  4. A set of “codes” was devised here, too, in order to generate a standardized dataset. The following four codes were adopted: (1) “c” for committing war crimes; (2) “o” for ordering and authorizing the commission of war crimes; (3) “f” for failure to take steps to stop or prevent war crimes; and (4) “p” for “permitting” the commission of war crimes, which is an ill-defined concept of liability but frequently used at post-WWII Allied war crimes trials.