Once you have access to Journey Insights Hub and a workspace, the next step is to prepare the text data you want to analyze.
Your dataset should include the open-ended feedback you want Journey Insights Hub to evaluate, such as survey responses, reviews, support tickets, or chat transcripts. You can also include metadata to provide additional context, such as dates, scores, customer segments, products, or regions.
Metadata allows you to filter your text analysis project and compare different subsets of records. Numerical fields can also be used in the Drivers feature to identify concepts associated with higher or lower scores.
Key vocabulary
Before preparing your dataset, it is helpful to understand a few commonly used terms.
Verbatim
The open-ended or conversational text that you want Journey Insights Hub to analyze
Record
A single row of source data containing a verbatim and any associated metadata
Metadata
Structured information that provides context for a record. Metadata may include dates, scores, regions, products, customer segments, or feedback channels
CSV file
A comma-separated values file used to organize data in a plain-text format. Most spreadsheet applications allow you to export a file in CSV format.
Supported file types
File type | Description |
CSV (.csv) | A comma-separated values file. Recommended for most uploads and large datasets because it provides a simple, consistent data structure. |
Excel (.xlsx or .xls) | A Microsoft Excel worksheet. Recommended when your data is organized in a table and you want to upload it directly from Excel. |
TSV (.tsv) | A tab-separated values file. It is structured similarly to a CSV file, but uses tabs to separate values. |
Recommendation: Use CSV when for large datasets or when you experience formatting issues with an Excel workbook.
Structure your dataset
Whether you upload a CSV, TSV, or Excel file, your dataset should follow the same basic structure:
The first row must contain column headers
Each additional row should represent one record
The file must contain at least one Text field
Each column should contain one consistent type of data
Dates, numbers and scores should use a consistent format throughout their columns
The dataset should not contain merged cells or additional headings
Prepare open-ended text for analysis
Your source file can contain more than one column of open-ended text. When preparing the data for a text analysis project, you can select the open-ended text columns you want to analyze and combine them into a single Text field.
The text analysis project uses one Text field, during project setup, Journey Insights Hub provides the option to select the open-ended text columns for analysis and combines them within the platform into one Text field for each record.
Each Text record may contain up to 500,000 characters. Other columns can be retained as metadata to provide context for the feedback.
Data fields
The first row of your dataset contains the column headers. During upload, each column must be assigned a data type.
You can define a column’s data type in either of two ways:
Add the data type before the column name in your source file, separated by an underscore. For example:
score_Satisfactionorstring_Region.Select or change the data type for each column during the upload process in Journey Insights Hub.
Data type | Examples |
• Text (Required) | Column header: |
• Contains the natural language feedback for Journey Insights Hub to analyze • There must be one and only one Text column per file • Each Text value may not exceed 500,000 characters | • The verification process took too long, and I had to request the code several times • The representative was helpful, but I waited several days for a response • The new dashboard is easier to navigate and gives me the information I need |
Title | Column header: |
• A short way to identify each record • The Title is not analyzed as part of the verbatim, but it can help organize records • Only one Title column is permitted per data file • Titles are optional | • Verification issue • Delayed support response • Positive dashboard feedback |
String | Column header: |
• Words or labels that help categorize records • String fields can be used to filter data when they contain up to 10,000 unique values • Include as many String columns as needed | Example:
• New customer • Enterprise • Returning customer |
Number | Column header: |
• Contains numerical data • Can optionally be used in the Drivers feature • Include as many Number columns as needed | Example:
• 1 • 4 • 12 |
Score | Column header: |
• Contains score or rating data • Scores should represent values where higher numbers indicate better results • Recommended for use with the Drivers feature • Include as many Score columns as needed | Example:
• 10 • 7 • 4.5 |
Date | Column header: |
• Contains dates or times • Accepts ISO 8601 strings, Unix timestamps, or US-style formats • Journey Insights Hub assumes dates are in UTC unless an ISO 8601 date includes a specific time zone • Helps you filter records and compare data across time periods • Include as many Date columns as needed | Example:
ISO 8601-formatted dates: • • •
US-style dates: • • • |
Formatting metadata with multiple values
A metadata field may sometimes contain more than one value for a single record.
For example, a survey may ask respondents which channels they used to contact support. A respondent might select email, chat and phone.
You can format multiple values in either of the following ways:
Use multiple columns with the same header
Create multiple columns with the same column name and enter one value in each cell.
string_SupportChannels | string_SupportChannels | string_SupportChannels |
Chat | Phone |
Some cells may be left blank when a record contains fewer values. In either format, Journey Insights Hub creates one metadata field containing multiple values.
Preparing multilingual datasets
Journey Insights Hub can analyze text natively in 14 supported languages. For the best results, separate multilingual data into one language per upload file. Upload and analyze each language as its own text analysis project so that the language model can learn the terminology and relationships within that language.