This is particularly relevant in our domain of interest (enterprise AI and specifically customer support), where Process Owners may know which dimensions to prioritize (inclusion, e.g., customer-reported symptom or severity), or know which aspects are irrelevant to their use case (exclusion, e.g. programming language or system). Our methods enable these preferences to be reflected in the resulting clusters.

Our use of prefixing is different (and, to our knowledge, novel): rather than aiming to improve performance on a given ground-truth label or metric, we seek to reshape the embedding space to discover alternative ways to cluster the same data. We achieve this by systematically changing the prefix (e.g., "Cluster by topic" vs. "Cluster by sentiment").

we leverage recent advances in instruction-tuned LLMs to manipulate the embedding space itself via prompting, while keeping the clustering algorithms unchanged.

We therefore argue that unsupervised clustering is an ill-defined problem, and that clustering is inherently subjective. In a space of many possible "correct" groupings, a generic clustering algorithm is unlikely to organize data in a way relevant to a user's specific use case.

Clustering aims to group similar data points; however, similarity for data points often depends on the analyst's perspective. Consider a dataset of social media posts: a content moderator might want to group posts with similar emotional tones and obscene language indicators, whereas a medical researcher might look for groupings based on mentions of drug use or illnesses, irrespective of sentiment.

Clustering is a central component of enterprise process analysis. For example, in IT Service Management (ITSM), common asks by Process Owners (POs) include "What are the most common user complaints?", "What are the underlying causes of problems?", and "Why are tickets rerouted to agents?". The answer POs seek involve grouping complex information into buckets that make sense from one's analysis perspective.

Instance methods Instances of Models are documents. Documents have many of their own built-in instance methods. We may also define our own custom document instance methods. // define a schema const animalSchema = new Schema({ name: String, type: String }, { // Assign a function to the "methods" object of our animalSchema through schema options. // By following this approach, there is no need to create a separate TS type to define the type of the instance functions. methods: { findSimilarTypes(cb) { return mongoose.model('Animal').find({ type: this.type }, cb); } } }); // Or, assign a function to the "methods" object of our animalSchema animalSchema.methods.findSimilarTypes = function(cb) { return mongoose.model('Animal').find({ type: this.type }, cb); }; Now all of our animal instances have a findSimilarTypes method available to them. const Animal = mongoose.model('Animal', animalSchema); const dog = new Animal({ type: 'dog' }); dog.findSimilarTypes((err, dogs) => { console.log(dogs); // woof }); Overwriting a default mongoose document method may lead to unpredictable results. See this for more details. The example above uses the Schema.methods object directly to save an instance method. You can also use the Schema.method() helper as described here. Do not declare methods using ES6 arrow functions (=>). Arrow functions explicitly prevent binding this, so your method will not have access to the document and the above examples will not work.

For example, what if your site has a customer interface and an “admin” interface? If the two have totally different designs and features, then it might be considerable overhead to ship the entirety of the admin interface to every customer on the regular site.

So, the sad conclusion is: all problems must be resolved individually depending on a specific usage context. There’s no silver bullet to resolve all ZeroDivisionErrors once and for all. And again, I am not even covering complex IO flows with retry policies and expotential timeouts.