AI clause training guidelines

For best results, it is important to select relevant text when training the HighQ AI to recognise custom clauses. Training with the correct text improves the results from the model.

Examples of good AI training practice

1. Improve results from the model by selecting entire paragraphs, as shown below.

2. If the clause contains multiple paragraphs, divide the clause into individual paragraphs for training.

3. If a paragraph is split across two pages, as shown below, select both halves as a single paragraph.

4. If a document contains text formatted in a table, select only the paragraph text, not the table row or cell.

5. If selecting an entire paragraph does not make sense, choose the part of the paragraph that is relevant to the clause being trained.

Examples of bad AI training practice

1. Do not select footers or other sections of text not related to the clause.

2. Do not select tables of content, an index or similar lists of text.

3. A short, one-line sentence does not provide enough information to train a clause.

4. Do not select short sections with just dates, names or section numbers.

5. Do not select sections that only contain numbers, such as tabulated data, as shown below.

6. Do not train the model with part of a sentence. Only select entire sentences

Note: Selecting multiple sections of repeated text does not improve the model. Once is sufficient.