Dialog tags

Tags are created and configured at the project level. Different projects can have different tags.

SpeechSense automatically detects whether the dialog contains specific key phrases or phrases with similar meaning and tags it. For example, there are tags that show the agent has thanked the customer for waiting or that the customer has contacted support for a second time.

To be assigned to a dialog, tags need to be activated. The activation takes place automatically as soon as you create a tag. With the tag activated, SpeechSense analyses the dialogs and assigns the tag if the dialog meets the specified conditions.

There are the following tag types in SpeechSense:

• Semantic tags rely on dialog analysis with the help of semantic attributes you address search queries to.
• Dictionary tags are based on searching for specific words and phrases in the dialogs.

For more info on creating, updating, and deleting tags, see these guides.

Where tags are displayedWhere tags are displayed

• On the Tags tab in the project.
• In reports:
  • In the dialog filters on the Dialogs tab.
  • In the evaluation parameter list.
• On the page of the dialog the tag is assigned to.

Semantic tagsSemantic tags

Semantic tags rely on dialog analysis with the help of semantic attributes. When creating a semantic tag, you select one of the semantic attributes and address a search query to it. SpeechSense analyses the dialogs and tags them if the dialog is semantically relevant to the search query in the context of the selected semantic attribute.

To manage tagging with more accuracy, set the minimum probability of the dialog satisfying the specified conditions. Only the dialogs for which that probability equals or exceeds the specified level will be tagged.

For example, you select Themes as the semantic attribute, Vacation in Sochi as your search query, and set 90% probability. SpeechSense will tag the dialogs with topics semantically related to Vacation in Sochi with a probability of at least 90%.

SpeechSense operates the following semantic attributes:

• Themes.
• Reasons.
• Results.
• Problems.
• Agent keywords.
• Client keywords.
• Common question: Preset questions of universal applicability.

By default, semantic tags get the Agent and Customer channels assigned at the same time, and this cannot be changed.

SuggestionsSuggestions

When creating a semantic tag, you can either enter your search query yourself or use suggestions. The available suggestions depend on the selected semantic attribute:

• For general questions, suggestions are preset and do not depend on what dialogs are selected for analysis.
• For all the other attributes, suggestions are dynamically generated for the selected dialogs. If you change your selection of dialogs for analysis, suggestions will be regenerated.

When generating suggestions, SpeechSense analyzes the selection and generates possible key phrases for the specified semantic attribute. This requires at least 100 dialogs; otherwise, SpeechSense will not be able to conduct a reliable analysis. There are no restrictions on the maximum number; however, if your selection contains more than 1,000 dialogs, SpeechSense will randomly select only 1,000 dialogs for analysis.

If there are more than 1,000 dialogs in the selection and you are not satisfied with the suggestions that were generated, click Refresh list for SpeechSense to re-select 1,000 dialogs at random and generate new suggestions.

Testing semantic tagsTesting semantic tags

When creating or editing a semantic tag, you can test it on a selection of dialogs. During the test, SpeechSense analyzes whether or not the dialog matches the conditions specified in the tag. The output will list the dialogs in descending order of match probability. If there are more than 1,000 dialogs in the selection, SpeechSense will select 1,000 dialogs covering the whole match probability range. During the test, you can change the semantic attribute, search query, or match probability value depending on the test results for more accurate tagging.

For more information about tag testing, see this guide.

Using semantic tags for dialog evaluationUsing semantic tags for dialog evaluation

You can use semantic tags to evaluate the agent's performance or the client's behavior, e.g., whether the agent was polite, whether the client was rude, etc. To do this, create a semantic tag with the relevant semantic attribute and search query. SpeechSense will assign this tag to the dialogs that meet this criterion.

For example, if looking for dialogs where the customer was rude with the agent, select the following when creating the tag:

• Semantic attribute: Common question.
• Search query: Was the client rude or aggressive at least once during the conversation?

SpeechSense will tag dialogs where the customer was rude or aggressive during the conversation.

Such use of semantic tags allows you to filter dialogs based on the agent's performance criteria and the client’s behavior, and use the same data as a filter or parameter in your reports.

Dictionary tagsDictionary tags

Dictionary tags are based on scanning dialogs for specified key phrases. You can either list the key phrases for tags or reuse them from dictionaries. Dictionaries contain a ready-made list of key phrases which you can use across multiple tags and manage from the same place.

To fine-tune key phrase detection:

• Enable the Exact match setting. In this case, related phrases will not be considered in key phrase detection.

• Set the distance between words. This setting determines the maximum allowable distance between words in a key phrase.

  For example, a tag is assigned the poor service key phrase. The search for this phrase will be performed differently depending on the Distance between words field value:

  Example in a dialog	Distance: 0	Distance: 1	Distance: 2
  poor service	found	found	found
  poor customer service	not found	found	found
  poor and slow service	not found	not found	found

In the dialog text, you can see the list of tags assigned for the agent and customer channels. If you click a tag, the words it refers to will be highlighted with the tag's color.

For more information about dictionary tags, see this guide.

Dictionary tag channelDictionary tag channel

Channel sets the part of the dialog SpeechSense will search for keywords in. You can use the following channels:

• Agent and customer: Whole dialog.
• Agent: Agent's utterances or messages only.
• Client: Customer's utterances or messages only.

Depending on the channel, SpeechSense looks for keywords in different parts of the dialog.

For example, the tag will not be assigned if it has the Client channel assigned, and its keyword only occurs in the agent's speech or messages.

The channel must be configured at the project level. This enables flexible tag assignment management across different projects. For more information about assigning a channel to a dictionary tag, see this guide.

Tag search area constraintsTag search area constraints

In SpeechSense, you can use constraints to narrow down the search area for dictionary tag keywords. If you set a constraint, a tag will trigger only if SpeechSense finds the keyword in the specified part of the dialog.

For example, you configure a constraint with the following settings:

• Constraint type: Start of dialog.
• Search direction: Forward.
• Limit the search area: Up to three utterances.

SpeechSense will only search for keywords in the first three utterances from the beginning of the dialog.

Every constraint in SpeechSense is associated with a particular event that marks the beginning of the search area. Available events:

• Beginning of the dialog.
• Long pause (pause length is measured in seconds and set manually).
• The agent interrupted the customer.
• The customer interrupted the agent.
• End of the dialog.
• Main tag triggered (for dependent tag).

By default, the maximum search area size is set to 1,000 utterances. To further refine the search area, specify the search direction from the event (forward, backward, or both) and the number of utterances to include in the search.

You can add multiple constraints to a single tag. The constraints sum up as per the logical AND operator.

Exclusion phrasesExclusion phrases

To increase a dictionary tag's accuracy, add exclusion phrases to it. If the dialog contains exclusion phrases, these will not trigger the tag.

You can add exclusion phrases manually or upload a phrase dictionary.

Let's say we create a tag called Fire and add the flamed keyword to it.
The tag will trigger on the phrase The battle flamed up; however, such a phrase does not indicate a fire. This is a false positive. To fix it, let's add the battle flamed up exclusion phrase.
Now the tag will not trigger on the phrase The battle flamed up but it will on The stove flamed up.

To fine-tune the exclusion phrase:

• Enable the Exact match setting. In this case, related phrases will not be considered in exclusion phrase detection.
• Set the distance between words. This setting determines the maximum allowable distance between words in an exclusion phrase.

Testing dictionary tagsTesting dictionary tags

When creating or editing a dictionary tag, you can test it on a selection of dialogs. During the test, SpeechSense will analyze the dialogs from the specified selection and list the dialogs eligible for the tag.

For more information about tag testing, see this guide.

Dependent tagsDependent tags

You can create a dependent tag for any SpeechSense tag. A dependent tag is one that triggers only if the main tag has done so. A main tag is one for which a dependent tag was created.

Both dictionary and semantic tags can act as basic and dependent tags, in any combination. Dependent tags can be nested down to any depth.

Dictionary tag dependency is implemented through the The main tag was triggered constraint automatically added to the dependent tag. You cannot delete this constraint. If you create a dependent dictionary tag for a semantic tag, you cannot set a constraint for this dictionary tag.

You can edit dependent tags in accordance with their type as basic semantic or dictionary tags.

When you delete the main tag, all its dependent tags will also be deleted.