How can I install html5lib on a dataproc cluster

Question

I have a dataproc pipeline with which I do webscraping and store data in gcp. Task setting is something like this:

    create_dataproc_cluster = DataprocCreateClusterOperator(
    task_id='create_dataproc_cluster',
    impersonation_chain=default_args['service_account'],
    cluster_name=default_args['cluster_name'],
    cluster_config=default_args['cluster_config'],
    region=default_args['region'],
    project_id=default_args['project_id'],
    timeout=36000
)

load_file_from_web = DataprocSubmitPySparkJobOperator(
    task_id='load_file_from_web',
    cluster_name=default_args['cluster_name'],
    region=default_args['region'],
    main=default_args['load_from_web_file'], (a python code to do web scarping)
    project_id=default_args['project_id'],
    labels={'team': 'AAA',
            'purpose': 'load_file_from_web'}
)
a task to delete the cluster like delete_dataproc_cluster 
start >> create_dataproc_cluster >> load_file_from_web >> delete_dataproc_cluster >> end

I also pull config and args from a json file:

 "cluster_config": {

    "config_bucket": "AAAA",
    "temp_bucket": "BBBB",
    "gce_cluster_config": {
      "internal_ip_only": true,
      "subnetwork_uri": "CCC",
      "service_account": "DDDD",
      "service_account_scopes": [
        "https://www...."
      ],
      "tags": [
        "JJJJ",
....
      ]
    },
    "master_config": {
      "image_uri":"a custom image",
      "num_instances": 1,
      "machine_type_uri": "..",
      "disk_config": {
        "boot_disk_type": "...",
        "boot_disk_size_gb": ...
      }
    },
    "worker_config": {
      "num_instances": 2,
      "machine_type_uri": "...",
      "disk_config": {
        "boot_disk_type": "...",
        "boot_disk_size_gb": ...
      }
    }
  }

However I got this error:

> Job failed with message [ImportError: html5lib not found, please
> install it]. Additional details can be found at:\nhttps:

Which one is the feasible solution and how can I implement it? Install the package in the cluster OR add init-action OR force the dataproc job to install the package before going through the main lines of the code?(install within the python code)