Budget Argument Mining

Background

Politics has the role of preparing a budget that takes into account income and expenditure and decides how to spend the money. The national budget is prepared by the Cabinet, and based on the draft budget, it is debated in the Diet before being approved. The budget of the local government is passed after the budget is prepared by the governor or the mayor and deliberated in the assembly. However, there is a problem that it is difficult to understand the background of the budget and the discussions that have led to the passage of the budget.

Purpose

Budget Argument Mining aims to connect published budget documents with the discussions included in the meeting minutes. Specifically, when a budget item (Amount, name of competent ministry/department, and explanation) is given, it finds politicians’ statements related to the budget (statements including the expression of the amount of money) in the Congressional Record and assigns three discussion labels [Claim, Premise, Other].

• Input
  • Budget Items (Budget ID, amount, name of competent ministry/department, description)
  • Minutes of the Assembly (Congress name, speaker, speech, monetary expression, relevant ID, argumentClass)

• Output
  • Congressional Record – Monetary Value Granted by Unique Representation Extractor (GINZA 4.0) – Grant the Budget ID of the Budget Item to the Related ID
  • argumentClass ( Seven classes )
    1. Premise : Past and Decisions
    2. Premise : Current and Future / Estimates
    3. Premise : Other：
    4. Claim : Opinions, suggestions, and questions:
    5. Claim : Other
    6. It is not a monetary expression.
    7. Other

• Evaluation
  • argumentClass
    – Correct answer rate = Correct answer question label ➗ Argument label
  • relatedID
    – Recall = number of correct answers in output ➗ number of correct answers
    – Precision = number of correct answers in output ➗ number of outputs (concatenation)

Annotation

How do we assign an argumentClass?

ArgumentClass is divided into “Claim,” “Premise,” “Not a monetary expression,” and “Other.

Claim (claim) summarizes claims by expression of opinions, proposals, and questions, and defines the rest as “others”. Premise is divided into “past” and “future” on the time axis, and “other” is defined as the case where neither is applicable.

What kind of argumentClass are there?

1. Premise : Past and Decisions
2. Premise : Current and Future / Estimates
3. Premise : Other：
4. Claim : Opinions, suggestions, and questions:
5. Claim : Other
6. It is not a monetary expression.
7. Other

argumentClass is attached to a “clause” , which includes a money expression.

What is “relatedID”?

The relatedID is assigned to the monetary expression contained in the statement in the meeting minutes. It is associated with the budget item (budgetId) that is related to the monetary expression included in the meeting minutes.

What is the data?

Diet and the House of Representatives and three local governments

• Diet and the House of Representatives　
• 012033 Otaru-city, Hokkaido
• 080004 Ibaraki
• 401307 Fukuoka-city, Fukuoka

対象とする会議録と予算

• Diet and the House of Representatives　
• 012033 Otaru-city, Hokkaido
  • 2019 ( before Corona ) , the first regular meeting in 2019
    • Minutes > The first regular meeting（2,743KB）
    • Budget > Overview of budget（PDF:947KB）
  • 2020 ( after Corona ) , the first regular meeting in 2020
    • Minutes > The first regular meeting（6,097KB）
    • Budget > Overview of budget（PDF:953KB）

• 080004 Ibaraki
• 401307 Fukuoka-city, Fukuoka

Budget field name

• In the case of Fukuoka City, the April 2020 supplementary budget is considered to be the first supplementary budget and will be “01”.

予算の json

[
    {
        "budgetId": "ID-2019-012033-00-000001",
        "budgetTitle": "平成３１年度当初予算案",
        "url": "https://www.city.otaru.lg.jp/docs/2020121200484/file_contents/03_H31_syuyo_zigyo_sashikae_r01.pdf",
        "budgetItem": "病児保育事業費補助金",
        "budget": "4,275千円",
        "categories": [
            "保育環境の整備　《公約1-①》",
            "安心して子どもを生み育てることのできるまち（子ども・子育て）"
        ],
        "typesOfAccount": null,
        "department": "福祉部",
        "budgetLastYear": null,
        "description": "市内の保育所・幼稚園・認定こども園、小学校に通う1歳6か月から小学校2年生までの児童を対象に、病気になった際の一時保育を実施する学校法人小樽学園いなほ幼稚園に対する補助",
        "budgetDifference": null
    }
]

会議録のフィールド名

会議録　json

[
    {
        "date": "2019-02-20",
        "localGovernmentCode": "011002",
        "localGovernmentName": "小樽市",
        "proceedingTitle": "平成３１年第１回定例会 本会議録",
        "url": "http://",
        "proceeding": [
            {
                "speakerPosition": "市長",
                "speaker": "迫 俊哉",
                "utterance": "平成 31 年第１回定例会の開会に当たり、ただいま上程されました各案件に係る提案理由の説明に先立ち、幾つか申し述べさせていただきます。まず、歳入についてでありますが、市税につきましては、個人市民税、法人市民税などで減収が見込まれるものの、固定資産税、都市計画税などで増収が見込まれることから、2.7％、３億 5,280万円増の135億 7,350万円を見込みました。",
                "moneyExpressions": [
                    {
                        "moneyExpression": "３億 5,280万円増",
                        "relatedID": null,
                        "argumentClass": "Premise : 未来（現在以降）・見積"
                    },
                    {
                        "moneyExpression": "135億 7,350万円",
                        "relatedID": null,
                        "argumentClass": "Premise : 未来（現在以降）・見積"
                    }
                ]
            }
        ]
    }
]

参考文献

国会および地方議会の予算審議を対象としたBudget Argument Miningタスクの提案  永渕景祐, 土屋彩夏, 木村泰知 言語処理学会第27回年次大会（NLP2021） P6-13 2021年3月